xref: /linux/include/net/tcp.h (revision 2624f124b3b5d550ab2fbef7ee3bc0e1fed09722)
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 TCP_DEBUG 1
22 #define FASTRETRANS_DEBUG 1
23 
24 #include <linux/config.h>
25 #include <linux/list.h>
26 #include <linux/tcp.h>
27 #include <linux/slab.h>
28 #include <linux/cache.h>
29 #include <linux/percpu.h>
30 
31 #include <net/inet_connection_sock.h>
32 #include <net/inet_timewait_sock.h>
33 #include <net/inet_hashtables.h>
34 #include <net/checksum.h>
35 #include <net/request_sock.h>
36 #include <net/sock.h>
37 #include <net/snmp.h>
38 #include <net/ip.h>
39 #include <net/tcp_states.h>
40 
41 #include <linux/seq_file.h>
42 
43 extern struct inet_hashinfo tcp_hashinfo;
44 
45 extern atomic_t tcp_orphan_count;
46 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
47 
48 #define MAX_TCP_HEADER	(128 + MAX_HEADER)
49 
50 /*
51  * Never offer a window over 32767 without using window scaling. Some
52  * poor stacks do signed 16bit maths!
53  */
54 #define MAX_TCP_WINDOW		32767U
55 
56 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
57 #define TCP_MIN_MSS		88U
58 
59 /* Minimal RCV_MSS. */
60 #define TCP_MIN_RCVMSS		536U
61 
62 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
63 #define TCP_FASTRETRANS_THRESH 3
64 
65 /* Maximal reordering. */
66 #define TCP_MAX_REORDERING	127
67 
68 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
69 #define TCP_MAX_QUICKACKS	16U
70 
71 /* urg_data states */
72 #define TCP_URG_VALID	0x0100
73 #define TCP_URG_NOTYET	0x0200
74 #define TCP_URG_READ	0x0400
75 
76 #define TCP_RETR1	3	/*
77 				 * This is how many retries it does before it
78 				 * tries to figure out if the gateway is
79 				 * down. Minimal RFC value is 3; it corresponds
80 				 * to ~3sec-8min depending on RTO.
81 				 */
82 
83 #define TCP_RETR2	15	/*
84 				 * This should take at least
85 				 * 90 minutes to time out.
86 				 * RFC1122 says that the limit is 100 sec.
87 				 * 15 is ~13-30min depending on RTO.
88 				 */
89 
90 #define TCP_SYN_RETRIES	 5	/* number of times to retry active opening a
91 				 * connection: ~180sec is RFC minumum	*/
92 
93 #define TCP_SYNACK_RETRIES 5	/* number of times to retry passive opening a
94 				 * connection: ~180sec is RFC minumum	*/
95 
96 
97 #define TCP_ORPHAN_RETRIES 7	/* number of times to retry on an orphaned
98 				 * socket. 7 is ~50sec-16min.
99 				 */
100 
101 
102 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
103 				  * state, about 60 seconds	*/
104 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
105                                  /* BSD style FIN_WAIT2 deadlock breaker.
106 				  * It used to be 3min, new value is 60sec,
107 				  * to combine FIN-WAIT-2 timeout with
108 				  * TIME-WAIT timer.
109 				  */
110 
111 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
112 #if HZ >= 100
113 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
114 #define TCP_ATO_MIN	((unsigned)(HZ/25))
115 #else
116 #define TCP_DELACK_MIN	4U
117 #define TCP_ATO_MIN	4U
118 #endif
119 #define TCP_RTO_MAX	((unsigned)(120*HZ))
120 #define TCP_RTO_MIN	((unsigned)(HZ/5))
121 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value	*/
122 
123 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
124 					                 * for local resources.
125 					                 */
126 
127 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
128 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
129 #define TCP_KEEPALIVE_INTVL	(75*HZ)
130 
131 #define MAX_TCP_KEEPIDLE	32767
132 #define MAX_TCP_KEEPINTVL	32767
133 #define MAX_TCP_KEEPCNT		127
134 #define MAX_TCP_SYNCNT		127
135 
136 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
137 #define TCP_SYNQ_HSIZE		512	/* Size of SYNACK hash table */
138 
139 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
140 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
141 					 * after this time. It should be equal
142 					 * (or greater than) TCP_TIMEWAIT_LEN
143 					 * to provide reliability equal to one
144 					 * provided by timewait state.
145 					 */
146 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
147 					 * timestamps. It must be less than
148 					 * minimal timewait lifetime.
149 					 */
150 /*
151  *	TCP option
152  */
153 
154 #define TCPOPT_NOP		1	/* Padding */
155 #define TCPOPT_EOL		0	/* End of options */
156 #define TCPOPT_MSS		2	/* Segment size negotiating */
157 #define TCPOPT_WINDOW		3	/* Window scaling */
158 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
159 #define TCPOPT_SACK             5       /* SACK Block */
160 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
161 
162 /*
163  *     TCP option lengths
164  */
165 
166 #define TCPOLEN_MSS            4
167 #define TCPOLEN_WINDOW         3
168 #define TCPOLEN_SACK_PERM      2
169 #define TCPOLEN_TIMESTAMP      10
170 
171 /* But this is what stacks really send out. */
172 #define TCPOLEN_TSTAMP_ALIGNED		12
173 #define TCPOLEN_WSCALE_ALIGNED		4
174 #define TCPOLEN_SACKPERM_ALIGNED	4
175 #define TCPOLEN_SACK_BASE		2
176 #define TCPOLEN_SACK_BASE_ALIGNED	4
177 #define TCPOLEN_SACK_PERBLOCK		8
178 
179 /* Flags in tp->nonagle */
180 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
181 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
182 #define TCP_NAGLE_PUSH		4	/* Cork is overriden for already queued data */
183 
184 extern struct inet_timewait_death_row tcp_death_row;
185 
186 /* sysctl variables for tcp */
187 extern int sysctl_tcp_timestamps;
188 extern int sysctl_tcp_window_scaling;
189 extern int sysctl_tcp_sack;
190 extern int sysctl_tcp_fin_timeout;
191 extern int sysctl_tcp_keepalive_time;
192 extern int sysctl_tcp_keepalive_probes;
193 extern int sysctl_tcp_keepalive_intvl;
194 extern int sysctl_tcp_syn_retries;
195 extern int sysctl_tcp_synack_retries;
196 extern int sysctl_tcp_retries1;
197 extern int sysctl_tcp_retries2;
198 extern int sysctl_tcp_orphan_retries;
199 extern int sysctl_tcp_syncookies;
200 extern int sysctl_tcp_retrans_collapse;
201 extern int sysctl_tcp_stdurg;
202 extern int sysctl_tcp_rfc1337;
203 extern int sysctl_tcp_abort_on_overflow;
204 extern int sysctl_tcp_max_orphans;
205 extern int sysctl_tcp_fack;
206 extern int sysctl_tcp_reordering;
207 extern int sysctl_tcp_ecn;
208 extern int sysctl_tcp_dsack;
209 extern int sysctl_tcp_mem[3];
210 extern int sysctl_tcp_wmem[3];
211 extern int sysctl_tcp_rmem[3];
212 extern int sysctl_tcp_app_win;
213 extern int sysctl_tcp_adv_win_scale;
214 extern int sysctl_tcp_tw_reuse;
215 extern int sysctl_tcp_frto;
216 extern int sysctl_tcp_low_latency;
217 extern int sysctl_tcp_nometrics_save;
218 extern int sysctl_tcp_moderate_rcvbuf;
219 extern int sysctl_tcp_tso_win_divisor;
220 
221 extern atomic_t tcp_memory_allocated;
222 extern atomic_t tcp_sockets_allocated;
223 extern int tcp_memory_pressure;
224 
225 /*
226  *	Pointers to address related TCP functions
227  *	(i.e. things that depend on the address family)
228  */
229 
230 struct tcp_func {
231 	int			(*queue_xmit)		(struct sk_buff *skb,
232 							 int ipfragok);
233 
234 	void			(*send_check)		(struct sock *sk,
235 							 struct tcphdr *th,
236 							 int len,
237 							 struct sk_buff *skb);
238 
239 	int			(*rebuild_header)	(struct sock *sk);
240 
241 	int			(*conn_request)		(struct sock *sk,
242 							 struct sk_buff *skb);
243 
244 	struct sock *		(*syn_recv_sock)	(struct sock *sk,
245 							 struct sk_buff *skb,
246 							 struct request_sock *req,
247 							 struct dst_entry *dst);
248 
249 	int			(*remember_stamp)	(struct sock *sk);
250 
251 	__u16			net_header_len;
252 
253 	int			(*setsockopt)		(struct sock *sk,
254 							 int level,
255 							 int optname,
256 							 char __user *optval,
257 							 int optlen);
258 
259 	int			(*getsockopt)		(struct sock *sk,
260 							 int level,
261 							 int optname,
262 							 char __user *optval,
263 							 int __user *optlen);
264 
265 
266 	void			(*addr2sockaddr)	(struct sock *sk,
267 							 struct sockaddr *);
268 
269 	int sockaddr_len;
270 };
271 
272 /*
273  * The next routines deal with comparing 32 bit unsigned ints
274  * and worry about wraparound (automatic with unsigned arithmetic).
275  */
276 
277 static inline int before(__u32 seq1, __u32 seq2)
278 {
279         return (__s32)(seq1-seq2) < 0;
280 }
281 
282 static inline int after(__u32 seq1, __u32 seq2)
283 {
284 	return (__s32)(seq2-seq1) < 0;
285 }
286 
287 
288 /* is s2<=s1<=s3 ? */
289 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
290 {
291 	return seq3 - seq2 >= seq1 - seq2;
292 }
293 
294 
295 extern struct proto tcp_prot;
296 
297 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
298 #define TCP_INC_STATS(field)		SNMP_INC_STATS(tcp_statistics, field)
299 #define TCP_INC_STATS_BH(field)		SNMP_INC_STATS_BH(tcp_statistics, field)
300 #define TCP_INC_STATS_USER(field) 	SNMP_INC_STATS_USER(tcp_statistics, field)
301 #define TCP_DEC_STATS(field)		SNMP_DEC_STATS(tcp_statistics, field)
302 #define TCP_ADD_STATS_BH(field, val)	SNMP_ADD_STATS_BH(tcp_statistics, field, val)
303 #define TCP_ADD_STATS_USER(field, val)	SNMP_ADD_STATS_USER(tcp_statistics, field, val)
304 
305 extern void			tcp_v4_err(struct sk_buff *skb, u32);
306 
307 extern void			tcp_shutdown (struct sock *sk, int how);
308 
309 extern int			tcp_v4_rcv(struct sk_buff *skb);
310 
311 extern int			tcp_v4_remember_stamp(struct sock *sk);
312 
313 extern int		    	tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
314 
315 extern int			tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
316 					    struct msghdr *msg, size_t size);
317 extern ssize_t			tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
318 
319 extern int			tcp_ioctl(struct sock *sk,
320 					  int cmd,
321 					  unsigned long arg);
322 
323 extern int			tcp_rcv_state_process(struct sock *sk,
324 						      struct sk_buff *skb,
325 						      struct tcphdr *th,
326 						      unsigned len);
327 
328 extern int			tcp_rcv_established(struct sock *sk,
329 						    struct sk_buff *skb,
330 						    struct tcphdr *th,
331 						    unsigned len);
332 
333 extern void			tcp_rcv_space_adjust(struct sock *sk);
334 
335 static inline void tcp_dec_quickack_mode(struct sock *sk,
336 					 const unsigned int pkts)
337 {
338 	struct inet_connection_sock *icsk = inet_csk(sk);
339 
340 	if (icsk->icsk_ack.quick) {
341 		if (pkts >= icsk->icsk_ack.quick) {
342 			icsk->icsk_ack.quick = 0;
343 			/* Leaving quickack mode we deflate ATO. */
344 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
345 		} else
346 			icsk->icsk_ack.quick -= pkts;
347 	}
348 }
349 
350 extern void tcp_enter_quickack_mode(struct sock *sk);
351 
352 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
353 {
354  	rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
355 }
356 
357 enum tcp_tw_status
358 {
359 	TCP_TW_SUCCESS = 0,
360 	TCP_TW_RST = 1,
361 	TCP_TW_ACK = 2,
362 	TCP_TW_SYN = 3
363 };
364 
365 
366 extern enum tcp_tw_status	tcp_timewait_state_process(struct inet_timewait_sock *tw,
367 							   struct sk_buff *skb,
368 							   const struct tcphdr *th);
369 
370 extern struct sock *		tcp_check_req(struct sock *sk,struct sk_buff *skb,
371 					      struct request_sock *req,
372 					      struct request_sock **prev);
373 extern int			tcp_child_process(struct sock *parent,
374 						  struct sock *child,
375 						  struct sk_buff *skb);
376 extern void			tcp_enter_frto(struct sock *sk);
377 extern void			tcp_enter_loss(struct sock *sk, int how);
378 extern void			tcp_clear_retrans(struct tcp_sock *tp);
379 extern void			tcp_update_metrics(struct sock *sk);
380 
381 extern void			tcp_close(struct sock *sk,
382 					  long timeout);
383 extern unsigned int		tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
384 
385 extern int			tcp_getsockopt(struct sock *sk, int level,
386 					       int optname,
387 					       char __user *optval,
388 					       int __user *optlen);
389 extern int			tcp_setsockopt(struct sock *sk, int level,
390 					       int optname, char __user *optval,
391 					       int optlen);
392 extern void			tcp_set_keepalive(struct sock *sk, int val);
393 extern int			tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
394 					    struct msghdr *msg,
395 					    size_t len, int nonblock,
396 					    int flags, int *addr_len);
397 
398 extern void			tcp_parse_options(struct sk_buff *skb,
399 						  struct tcp_options_received *opt_rx,
400 						  int estab);
401 
402 /*
403  *	TCP v4 functions exported for the inet6 API
404  */
405 
406 extern void		       	tcp_v4_send_check(struct sock *sk,
407 						  struct tcphdr *th, int len,
408 						  struct sk_buff *skb);
409 
410 extern int			tcp_v4_conn_request(struct sock *sk,
411 						    struct sk_buff *skb);
412 
413 extern struct sock *		tcp_create_openreq_child(struct sock *sk,
414 							 struct request_sock *req,
415 							 struct sk_buff *skb);
416 
417 extern struct sock *		tcp_v4_syn_recv_sock(struct sock *sk,
418 						     struct sk_buff *skb,
419 						     struct request_sock *req,
420 							struct dst_entry *dst);
421 
422 extern int			tcp_v4_do_rcv(struct sock *sk,
423 					      struct sk_buff *skb);
424 
425 extern int			tcp_v4_connect(struct sock *sk,
426 					       struct sockaddr *uaddr,
427 					       int addr_len);
428 
429 extern int			tcp_connect(struct sock *sk);
430 
431 extern struct sk_buff *		tcp_make_synack(struct sock *sk,
432 						struct dst_entry *dst,
433 						struct request_sock *req);
434 
435 extern int			tcp_disconnect(struct sock *sk, int flags);
436 
437 extern void			tcp_unhash(struct sock *sk);
438 
439 extern int			tcp_v4_hash_connecting(struct sock *sk);
440 
441 
442 /* From syncookies.c */
443 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
444 				    struct ip_options *opt);
445 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
446 				     __u16 *mss);
447 
448 /* tcp_output.c */
449 
450 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
451 				      unsigned int cur_mss, int nonagle);
452 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
453 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
454 extern void tcp_xmit_retransmit_queue(struct sock *);
455 extern void tcp_simple_retransmit(struct sock *);
456 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
457 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
458 
459 extern void tcp_send_probe0(struct sock *);
460 extern void tcp_send_partial(struct sock *);
461 extern int  tcp_write_wakeup(struct sock *);
462 extern void tcp_send_fin(struct sock *sk);
463 extern void tcp_send_active_reset(struct sock *sk,
464                                   unsigned int __nocast priority);
465 extern int  tcp_send_synack(struct sock *);
466 extern void tcp_push_one(struct sock *, unsigned int mss_now);
467 extern void tcp_send_ack(struct sock *sk);
468 extern void tcp_send_delayed_ack(struct sock *sk);
469 
470 /* tcp_input.c */
471 extern void tcp_cwnd_application_limited(struct sock *sk);
472 
473 /* tcp_timer.c */
474 extern void tcp_init_xmit_timers(struct sock *);
475 static inline void tcp_clear_xmit_timers(struct sock *sk)
476 {
477 	inet_csk_clear_xmit_timers(sk);
478 }
479 
480 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
481 extern unsigned int tcp_current_mss(struct sock *sk, int large);
482 
483 /* tcp.c */
484 extern void tcp_get_info(struct sock *, struct tcp_info *);
485 
486 /* Read 'sendfile()'-style from a TCP socket */
487 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
488 				unsigned int, size_t);
489 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
490 			 sk_read_actor_t recv_actor);
491 
492 /* Initialize RCV_MSS value.
493  * RCV_MSS is an our guess about MSS used by the peer.
494  * We haven't any direct information about the MSS.
495  * It's better to underestimate the RCV_MSS rather than overestimate.
496  * Overestimations make us ACKing less frequently than needed.
497  * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
498  */
499 
500 static inline void tcp_initialize_rcv_mss(struct sock *sk)
501 {
502 	struct tcp_sock *tp = tcp_sk(sk);
503 	unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
504 
505 	hint = min(hint, tp->rcv_wnd/2);
506 	hint = min(hint, TCP_MIN_RCVMSS);
507 	hint = max(hint, TCP_MIN_MSS);
508 
509 	inet_csk(sk)->icsk_ack.rcv_mss = hint;
510 }
511 
512 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
513 {
514 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
515 			       ntohl(TCP_FLAG_ACK) |
516 			       snd_wnd);
517 }
518 
519 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
520 {
521 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
522 }
523 
524 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
525 {
526 	if (skb_queue_empty(&tp->out_of_order_queue) &&
527 	    tp->rcv_wnd &&
528 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
529 	    !tp->urg_data)
530 		tcp_fast_path_on(tp);
531 }
532 
533 /* Compute the actual receive window we are currently advertising.
534  * Rcv_nxt can be after the window if our peer push more data
535  * than the offered window.
536  */
537 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
538 {
539 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
540 
541 	if (win < 0)
542 		win = 0;
543 	return (u32) win;
544 }
545 
546 /* Choose a new window, without checks for shrinking, and without
547  * scaling applied to the result.  The caller does these things
548  * if necessary.  This is a "raw" window selection.
549  */
550 extern u32	__tcp_select_window(struct sock *sk);
551 
552 /* TCP timestamps are only 32-bits, this causes a slight
553  * complication on 64-bit systems since we store a snapshot
554  * of jiffies in the buffer control blocks below.  We decidely
555  * only use of the low 32-bits of jiffies and hide the ugly
556  * casts with the following macro.
557  */
558 #define tcp_time_stamp		((__u32)(jiffies))
559 
560 /* This is what the send packet queueing engine uses to pass
561  * TCP per-packet control information to the transmission
562  * code.  We also store the host-order sequence numbers in
563  * here too.  This is 36 bytes on 32-bit architectures,
564  * 40 bytes on 64-bit machines, if this grows please adjust
565  * skbuff.h:skbuff->cb[xxx] size appropriately.
566  */
567 struct tcp_skb_cb {
568 	union {
569 		struct inet_skb_parm	h4;
570 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
571 		struct inet6_skb_parm	h6;
572 #endif
573 	} header;	/* For incoming frames		*/
574 	__u32		seq;		/* Starting sequence number	*/
575 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
576 	__u32		when;		/* used to compute rtt's	*/
577 	__u8		flags;		/* TCP header flags.		*/
578 
579 	/* NOTE: These must match up to the flags byte in a
580 	 *       real TCP header.
581 	 */
582 #define TCPCB_FLAG_FIN		0x01
583 #define TCPCB_FLAG_SYN		0x02
584 #define TCPCB_FLAG_RST		0x04
585 #define TCPCB_FLAG_PSH		0x08
586 #define TCPCB_FLAG_ACK		0x10
587 #define TCPCB_FLAG_URG		0x20
588 #define TCPCB_FLAG_ECE		0x40
589 #define TCPCB_FLAG_CWR		0x80
590 
591 	__u8		sacked;		/* State flags for SACK/FACK.	*/
592 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
593 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
594 #define TCPCB_LOST		0x04	/* SKB is lost			*/
595 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
596 
597 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
598 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
599 
600 #define TCPCB_URG		0x20	/* Urgent pointer advenced here	*/
601 
602 #define TCPCB_AT_TAIL		(TCPCB_URG)
603 
604 	__u16		urg_ptr;	/* Valid w/URG flags is set.	*/
605 	__u32		ack_seq;	/* Sequence number ACK'd	*/
606 };
607 
608 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
609 
610 #include <net/tcp_ecn.h>
611 
612 /* Due to TSO, an SKB can be composed of multiple actual
613  * packets.  To keep these tracked properly, we use this.
614  */
615 static inline int tcp_skb_pcount(const struct sk_buff *skb)
616 {
617 	return skb_shinfo(skb)->tso_segs;
618 }
619 
620 /* This is valid iff tcp_skb_pcount() > 1. */
621 static inline int tcp_skb_mss(const struct sk_buff *skb)
622 {
623 	return skb_shinfo(skb)->tso_size;
624 }
625 
626 static inline void tcp_dec_pcount_approx(__u32 *count,
627 					 const struct sk_buff *skb)
628 {
629 	if (*count) {
630 		*count -= tcp_skb_pcount(skb);
631 		if ((int)*count < 0)
632 			*count = 0;
633 	}
634 }
635 
636 static inline void tcp_packets_out_inc(struct sock *sk,
637 				       struct tcp_sock *tp,
638 				       const struct sk_buff *skb)
639 {
640 	int orig = tp->packets_out;
641 
642 	tp->packets_out += tcp_skb_pcount(skb);
643 	if (!orig)
644 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
645 					  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
646 }
647 
648 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
649 				       const struct sk_buff *skb)
650 {
651 	tp->packets_out -= tcp_skb_pcount(skb);
652 }
653 
654 /* Events passed to congestion control interface */
655 enum tcp_ca_event {
656 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
657 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
658 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
659 	CA_EVENT_FRTO,		/* fast recovery timeout */
660 	CA_EVENT_LOSS,		/* loss timeout */
661 	CA_EVENT_FAST_ACK,	/* in sequence ack */
662 	CA_EVENT_SLOW_ACK,	/* other ack */
663 };
664 
665 /*
666  * Interface for adding new TCP congestion control handlers
667  */
668 #define TCP_CA_NAME_MAX	16
669 struct tcp_congestion_ops {
670 	struct list_head	list;
671 
672 	/* initialize private data (optional) */
673 	void (*init)(struct sock *sk);
674 	/* cleanup private data  (optional) */
675 	void (*release)(struct sock *sk);
676 
677 	/* return slow start threshold (required) */
678 	u32 (*ssthresh)(struct sock *sk);
679 	/* lower bound for congestion window (optional) */
680 	u32 (*min_cwnd)(struct sock *sk);
681 	/* do new cwnd calculation (required) */
682 	void (*cong_avoid)(struct sock *sk, u32 ack,
683 			   u32 rtt, u32 in_flight, int good_ack);
684 	/* round trip time sample per acked packet (optional) */
685 	void (*rtt_sample)(struct sock *sk, u32 usrtt);
686 	/* call before changing ca_state (optional) */
687 	void (*set_state)(struct sock *sk, u8 new_state);
688 	/* call when cwnd event occurs (optional) */
689 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
690 	/* new value of cwnd after loss (optional) */
691 	u32  (*undo_cwnd)(struct sock *sk);
692 	/* hook for packet ack accounting (optional) */
693 	void (*pkts_acked)(struct sock *sk, u32 num_acked);
694 	/* get info for inet_diag (optional) */
695 	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
696 
697 	char 		name[TCP_CA_NAME_MAX];
698 	struct module 	*owner;
699 };
700 
701 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
702 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
703 
704 extern void tcp_init_congestion_control(struct sock *sk);
705 extern void tcp_cleanup_congestion_control(struct sock *sk);
706 extern int tcp_set_default_congestion_control(const char *name);
707 extern void tcp_get_default_congestion_control(char *name);
708 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
709 
710 extern struct tcp_congestion_ops tcp_init_congestion_ops;
711 extern u32 tcp_reno_ssthresh(struct sock *sk);
712 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
713 				u32 rtt, u32 in_flight, int flag);
714 extern u32 tcp_reno_min_cwnd(struct sock *sk);
715 extern struct tcp_congestion_ops tcp_reno;
716 
717 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
718 {
719 	struct inet_connection_sock *icsk = inet_csk(sk);
720 
721 	if (icsk->icsk_ca_ops->set_state)
722 		icsk->icsk_ca_ops->set_state(sk, ca_state);
723 	icsk->icsk_ca_state = ca_state;
724 }
725 
726 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
727 {
728 	const struct inet_connection_sock *icsk = inet_csk(sk);
729 
730 	if (icsk->icsk_ca_ops->cwnd_event)
731 		icsk->icsk_ca_ops->cwnd_event(sk, event);
732 }
733 
734 /* This determines how many packets are "in the network" to the best
735  * of our knowledge.  In many cases it is conservative, but where
736  * detailed information is available from the receiver (via SACK
737  * blocks etc.) we can make more aggressive calculations.
738  *
739  * Use this for decisions involving congestion control, use just
740  * tp->packets_out to determine if the send queue is empty or not.
741  *
742  * Read this equation as:
743  *
744  *	"Packets sent once on transmission queue" MINUS
745  *	"Packets left network, but not honestly ACKed yet" PLUS
746  *	"Packets fast retransmitted"
747  */
748 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
749 {
750 	return (tp->packets_out - tp->left_out + tp->retrans_out);
751 }
752 
753 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
754  * The exception is rate halving phase, when cwnd is decreasing towards
755  * ssthresh.
756  */
757 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
758 {
759 	const struct tcp_sock *tp = tcp_sk(sk);
760 	if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
761 		return tp->snd_ssthresh;
762 	else
763 		return max(tp->snd_ssthresh,
764 			   ((tp->snd_cwnd >> 1) +
765 			    (tp->snd_cwnd >> 2)));
766 }
767 
768 static inline void tcp_sync_left_out(struct tcp_sock *tp)
769 {
770 	if (tp->rx_opt.sack_ok &&
771 	    (tp->sacked_out >= tp->packets_out - tp->lost_out))
772 		tp->sacked_out = tp->packets_out - tp->lost_out;
773 	tp->left_out = tp->sacked_out + tp->lost_out;
774 }
775 
776 /* Set slow start threshold and cwnd not falling to slow start */
777 static inline void __tcp_enter_cwr(struct sock *sk)
778 {
779 	const struct inet_connection_sock *icsk = inet_csk(sk);
780 	struct tcp_sock *tp = tcp_sk(sk);
781 
782 	tp->undo_marker = 0;
783 	tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
784 	tp->snd_cwnd = min(tp->snd_cwnd,
785 			   tcp_packets_in_flight(tp) + 1U);
786 	tp->snd_cwnd_cnt = 0;
787 	tp->high_seq = tp->snd_nxt;
788 	tp->snd_cwnd_stamp = tcp_time_stamp;
789 	TCP_ECN_queue_cwr(tp);
790 }
791 
792 static inline void tcp_enter_cwr(struct sock *sk)
793 {
794 	struct tcp_sock *tp = tcp_sk(sk);
795 
796 	tp->prior_ssthresh = 0;
797 	if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {
798 		__tcp_enter_cwr(sk);
799 		tcp_set_ca_state(sk, TCP_CA_CWR);
800 	}
801 }
802 
803 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
804 
805 /* Slow start with delack produces 3 packets of burst, so that
806  * it is safe "de facto".
807  */
808 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
809 {
810 	return 3;
811 }
812 
813 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
814 					   const struct sk_buff *skb)
815 {
816 	if (skb->len < mss)
817 		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
818 }
819 
820 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
821 {
822 	const struct inet_connection_sock *icsk = inet_csk(sk);
823 	if (!tp->packets_out && !icsk->icsk_pending)
824 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
825 					  icsk->icsk_rto, TCP_RTO_MAX);
826 }
827 
828 static __inline__ void tcp_push_pending_frames(struct sock *sk,
829 					       struct tcp_sock *tp)
830 {
831 	__tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
832 }
833 
834 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
835 {
836 	tp->snd_wl1 = seq;
837 }
838 
839 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
840 {
841 	tp->snd_wl1 = seq;
842 }
843 
844 /*
845  * Calculate(/check) TCP checksum
846  */
847 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
848 				   unsigned long saddr, unsigned long daddr,
849 				   unsigned long base)
850 {
851 	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
852 }
853 
854 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
855 {
856 	return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
857 }
858 
859 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
860 {
861 	return skb->ip_summed != CHECKSUM_UNNECESSARY &&
862 		__tcp_checksum_complete(skb);
863 }
864 
865 /* Prequeue for VJ style copy to user, combined with checksumming. */
866 
867 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
868 {
869 	tp->ucopy.task = NULL;
870 	tp->ucopy.len = 0;
871 	tp->ucopy.memory = 0;
872 	skb_queue_head_init(&tp->ucopy.prequeue);
873 }
874 
875 /* Packet is added to VJ-style prequeue for processing in process
876  * context, if a reader task is waiting. Apparently, this exciting
877  * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
878  * failed somewhere. Latency? Burstiness? Well, at least now we will
879  * see, why it failed. 8)8)				  --ANK
880  *
881  * NOTE: is this not too big to inline?
882  */
883 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
884 {
885 	struct tcp_sock *tp = tcp_sk(sk);
886 
887 	if (!sysctl_tcp_low_latency && tp->ucopy.task) {
888 		__skb_queue_tail(&tp->ucopy.prequeue, skb);
889 		tp->ucopy.memory += skb->truesize;
890 		if (tp->ucopy.memory > sk->sk_rcvbuf) {
891 			struct sk_buff *skb1;
892 
893 			BUG_ON(sock_owned_by_user(sk));
894 
895 			while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
896 				sk->sk_backlog_rcv(sk, skb1);
897 				NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
898 			}
899 
900 			tp->ucopy.memory = 0;
901 		} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
902 			wake_up_interruptible(sk->sk_sleep);
903 			if (!inet_csk_ack_scheduled(sk))
904 				inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
905 						          (3 * TCP_RTO_MIN) / 4,
906 							  TCP_RTO_MAX);
907 		}
908 		return 1;
909 	}
910 	return 0;
911 }
912 
913 
914 #undef STATE_TRACE
915 
916 #ifdef STATE_TRACE
917 static const char *statename[]={
918 	"Unused","Established","Syn Sent","Syn Recv",
919 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
920 	"Close Wait","Last ACK","Listen","Closing"
921 };
922 #endif
923 
924 static __inline__ void tcp_set_state(struct sock *sk, int state)
925 {
926 	int oldstate = sk->sk_state;
927 
928 	switch (state) {
929 	case TCP_ESTABLISHED:
930 		if (oldstate != TCP_ESTABLISHED)
931 			TCP_INC_STATS(TCP_MIB_CURRESTAB);
932 		break;
933 
934 	case TCP_CLOSE:
935 		if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
936 			TCP_INC_STATS(TCP_MIB_ESTABRESETS);
937 
938 		sk->sk_prot->unhash(sk);
939 		if (inet_csk(sk)->icsk_bind_hash &&
940 		    !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
941 			inet_put_port(&tcp_hashinfo, sk);
942 		/* fall through */
943 	default:
944 		if (oldstate==TCP_ESTABLISHED)
945 			TCP_DEC_STATS(TCP_MIB_CURRESTAB);
946 	}
947 
948 	/* Change state AFTER socket is unhashed to avoid closed
949 	 * socket sitting in hash tables.
950 	 */
951 	sk->sk_state = state;
952 
953 #ifdef STATE_TRACE
954 	SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
955 #endif
956 }
957 
958 static __inline__ void tcp_done(struct sock *sk)
959 {
960 	tcp_set_state(sk, TCP_CLOSE);
961 	tcp_clear_xmit_timers(sk);
962 
963 	sk->sk_shutdown = SHUTDOWN_MASK;
964 
965 	if (!sock_flag(sk, SOCK_DEAD))
966 		sk->sk_state_change(sk);
967 	else
968 		inet_csk_destroy_sock(sk);
969 }
970 
971 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
972 {
973 	rx_opt->dsack = 0;
974 	rx_opt->eff_sacks = 0;
975 	rx_opt->num_sacks = 0;
976 }
977 
978 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
979 {
980 	if (tp->rx_opt.tstamp_ok) {
981 		*ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
982 					  (TCPOPT_NOP << 16) |
983 					  (TCPOPT_TIMESTAMP << 8) |
984 					  TCPOLEN_TIMESTAMP);
985 		*ptr++ = htonl(tstamp);
986 		*ptr++ = htonl(tp->rx_opt.ts_recent);
987 	}
988 	if (tp->rx_opt.eff_sacks) {
989 		struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
990 		int this_sack;
991 
992 		*ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
993 					  (TCPOPT_NOP << 16) |
994 					  (TCPOPT_SACK << 8) |
995 					  (TCPOLEN_SACK_BASE +
996 					   (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
997 		for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
998 			*ptr++ = htonl(sp[this_sack].start_seq);
999 			*ptr++ = htonl(sp[this_sack].end_seq);
1000 		}
1001 		if (tp->rx_opt.dsack) {
1002 			tp->rx_opt.dsack = 0;
1003 			tp->rx_opt.eff_sacks--;
1004 		}
1005 	}
1006 }
1007 
1008 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1009  * If this is every changed make sure to change the definition of
1010  * MAX_SYN_SIZE to match the new maximum number of options that you
1011  * can generate.
1012  */
1013 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1014 					     int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1015 {
1016 	/* We always get an MSS option.
1017 	 * The option bytes which will be seen in normal data
1018 	 * packets should timestamps be used, must be in the MSS
1019 	 * advertised.  But we subtract them from tp->mss_cache so
1020 	 * that calculations in tcp_sendmsg are simpler etc.
1021 	 * So account for this fact here if necessary.  If we
1022 	 * don't do this correctly, as a receiver we won't
1023 	 * recognize data packets as being full sized when we
1024 	 * should, and thus we won't abide by the delayed ACK
1025 	 * rules correctly.
1026 	 * SACKs don't matter, we never delay an ACK when we
1027 	 * have any of those going out.
1028 	 */
1029 	*ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1030 	if (ts) {
1031 		if(sack)
1032 			*ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1033 						  (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1034 		else
1035 			*ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1036 						  (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1037 		*ptr++ = htonl(tstamp);		/* TSVAL */
1038 		*ptr++ = htonl(ts_recent);	/* TSECR */
1039 	} else if(sack)
1040 		*ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1041 					  (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1042 	if (offer_wscale)
1043 		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1044 }
1045 
1046 /* Determine a window scaling and initial window to offer. */
1047 extern void tcp_select_initial_window(int __space, __u32 mss,
1048 				      __u32 *rcv_wnd, __u32 *window_clamp,
1049 				      int wscale_ok, __u8 *rcv_wscale);
1050 
1051 static inline int tcp_win_from_space(int space)
1052 {
1053 	return sysctl_tcp_adv_win_scale<=0 ?
1054 		(space>>(-sysctl_tcp_adv_win_scale)) :
1055 		space - (space>>sysctl_tcp_adv_win_scale);
1056 }
1057 
1058 /* Note: caller must be prepared to deal with negative returns */
1059 static inline int tcp_space(const struct sock *sk)
1060 {
1061 	return tcp_win_from_space(sk->sk_rcvbuf -
1062 				  atomic_read(&sk->sk_rmem_alloc));
1063 }
1064 
1065 static inline int tcp_full_space(const struct sock *sk)
1066 {
1067 	return tcp_win_from_space(sk->sk_rcvbuf);
1068 }
1069 
1070 static __inline__ void tcp_openreq_init(struct request_sock *req,
1071 					struct tcp_options_received *rx_opt,
1072 					struct sk_buff *skb)
1073 {
1074 	struct inet_request_sock *ireq = inet_rsk(req);
1075 
1076 	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
1077 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1078 	req->mss = rx_opt->mss_clamp;
1079 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1080 	ireq->tstamp_ok = rx_opt->tstamp_ok;
1081 	ireq->sack_ok = rx_opt->sack_ok;
1082 	ireq->snd_wscale = rx_opt->snd_wscale;
1083 	ireq->wscale_ok = rx_opt->wscale_ok;
1084 	ireq->acked = 0;
1085 	ireq->ecn_ok = 0;
1086 	ireq->rmt_port = skb->h.th->source;
1087 }
1088 
1089 extern void tcp_enter_memory_pressure(void);
1090 
1091 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1092 {
1093 	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1094 }
1095 
1096 static inline int keepalive_time_when(const struct tcp_sock *tp)
1097 {
1098 	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1099 }
1100 
1101 static inline int tcp_fin_time(const struct sock *sk)
1102 {
1103 	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1104 	const int rto = inet_csk(sk)->icsk_rto;
1105 
1106 	if (fin_timeout < (rto << 2) - (rto >> 1))
1107 		fin_timeout = (rto << 2) - (rto >> 1);
1108 
1109 	return fin_timeout;
1110 }
1111 
1112 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1113 {
1114 	if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1115 		return 0;
1116 	if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1117 		return 0;
1118 
1119 	/* RST segments are not recommended to carry timestamp,
1120 	   and, if they do, it is recommended to ignore PAWS because
1121 	   "their cleanup function should take precedence over timestamps."
1122 	   Certainly, it is mistake. It is necessary to understand the reasons
1123 	   of this constraint to relax it: if peer reboots, clock may go
1124 	   out-of-sync and half-open connections will not be reset.
1125 	   Actually, the problem would be not existing if all
1126 	   the implementations followed draft about maintaining clock
1127 	   via reboots. Linux-2.2 DOES NOT!
1128 
1129 	   However, we can relax time bounds for RST segments to MSL.
1130 	 */
1131 	if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1132 		return 0;
1133 	return 1;
1134 }
1135 
1136 #define TCP_CHECK_TIMER(sk) do { } while (0)
1137 
1138 static inline int tcp_use_frto(const struct sock *sk)
1139 {
1140 	const struct tcp_sock *tp = tcp_sk(sk);
1141 
1142 	/* F-RTO must be activated in sysctl and there must be some
1143 	 * unsent new data, and the advertised window should allow
1144 	 * sending it.
1145 	 */
1146 	return (sysctl_tcp_frto && sk->sk_send_head &&
1147 		!after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1148 		       tp->snd_una + tp->snd_wnd));
1149 }
1150 
1151 static inline void tcp_mib_init(void)
1152 {
1153 	/* See RFC 2012 */
1154 	TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1155 	TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1156 	TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1157 	TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1158 }
1159 
1160 /* /proc */
1161 enum tcp_seq_states {
1162 	TCP_SEQ_STATE_LISTENING,
1163 	TCP_SEQ_STATE_OPENREQ,
1164 	TCP_SEQ_STATE_ESTABLISHED,
1165 	TCP_SEQ_STATE_TIME_WAIT,
1166 };
1167 
1168 struct tcp_seq_afinfo {
1169 	struct module		*owner;
1170 	char			*name;
1171 	sa_family_t		family;
1172 	int			(*seq_show) (struct seq_file *m, void *v);
1173 	struct file_operations	*seq_fops;
1174 };
1175 
1176 struct tcp_iter_state {
1177 	sa_family_t		family;
1178 	enum tcp_seq_states	state;
1179 	struct sock		*syn_wait_sk;
1180 	int			bucket, sbucket, num, uid;
1181 	struct seq_operations	seq_ops;
1182 };
1183 
1184 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1185 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1186 
1187 extern struct request_sock_ops tcp_request_sock_ops;
1188 
1189 extern int tcp_v4_destroy_sock(struct sock *sk);
1190 
1191 #ifdef CONFIG_PROC_FS
1192 extern int  tcp4_proc_init(void);
1193 extern void tcp4_proc_exit(void);
1194 #endif
1195 
1196 extern void tcp_v4_init(struct net_proto_family *ops);
1197 extern void tcp_init(void);
1198 
1199 #endif	/* _TCP_H */
1200