xref: /linux/include/net/tcp.h (revision 54a8a2220c936a47840c9a3d74910c5a56fae2ed)
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, gfp_t priority);
464 extern int  tcp_send_synack(struct sock *);
465 extern void tcp_push_one(struct sock *, unsigned int mss_now);
466 extern void tcp_send_ack(struct sock *sk);
467 extern void tcp_send_delayed_ack(struct sock *sk);
468 
469 /* tcp_input.c */
470 extern void tcp_cwnd_application_limited(struct sock *sk);
471 
472 /* tcp_timer.c */
473 extern void tcp_init_xmit_timers(struct sock *);
474 static inline void tcp_clear_xmit_timers(struct sock *sk)
475 {
476 	inet_csk_clear_xmit_timers(sk);
477 }
478 
479 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
480 extern unsigned int tcp_current_mss(struct sock *sk, int large);
481 
482 /* tcp.c */
483 extern void tcp_get_info(struct sock *, struct tcp_info *);
484 
485 /* Read 'sendfile()'-style from a TCP socket */
486 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
487 				unsigned int, size_t);
488 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
489 			 sk_read_actor_t recv_actor);
490 
491 /* Initialize RCV_MSS value.
492  * RCV_MSS is an our guess about MSS used by the peer.
493  * We haven't any direct information about the MSS.
494  * It's better to underestimate the RCV_MSS rather than overestimate.
495  * Overestimations make us ACKing less frequently than needed.
496  * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
497  */
498 
499 static inline void tcp_initialize_rcv_mss(struct sock *sk)
500 {
501 	struct tcp_sock *tp = tcp_sk(sk);
502 	unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
503 
504 	hint = min(hint, tp->rcv_wnd/2);
505 	hint = min(hint, TCP_MIN_RCVMSS);
506 	hint = max(hint, TCP_MIN_MSS);
507 
508 	inet_csk(sk)->icsk_ack.rcv_mss = hint;
509 }
510 
511 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
512 {
513 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
514 			       ntohl(TCP_FLAG_ACK) |
515 			       snd_wnd);
516 }
517 
518 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
519 {
520 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
521 }
522 
523 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
524 {
525 	if (skb_queue_empty(&tp->out_of_order_queue) &&
526 	    tp->rcv_wnd &&
527 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
528 	    !tp->urg_data)
529 		tcp_fast_path_on(tp);
530 }
531 
532 /* Compute the actual receive window we are currently advertising.
533  * Rcv_nxt can be after the window if our peer push more data
534  * than the offered window.
535  */
536 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
537 {
538 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
539 
540 	if (win < 0)
541 		win = 0;
542 	return (u32) win;
543 }
544 
545 /* Choose a new window, without checks for shrinking, and without
546  * scaling applied to the result.  The caller does these things
547  * if necessary.  This is a "raw" window selection.
548  */
549 extern u32	__tcp_select_window(struct sock *sk);
550 
551 /* TCP timestamps are only 32-bits, this causes a slight
552  * complication on 64-bit systems since we store a snapshot
553  * of jiffies in the buffer control blocks below.  We decidely
554  * only use of the low 32-bits of jiffies and hide the ugly
555  * casts with the following macro.
556  */
557 #define tcp_time_stamp		((__u32)(jiffies))
558 
559 /* This is what the send packet queueing engine uses to pass
560  * TCP per-packet control information to the transmission
561  * code.  We also store the host-order sequence numbers in
562  * here too.  This is 36 bytes on 32-bit architectures,
563  * 40 bytes on 64-bit machines, if this grows please adjust
564  * skbuff.h:skbuff->cb[xxx] size appropriately.
565  */
566 struct tcp_skb_cb {
567 	union {
568 		struct inet_skb_parm	h4;
569 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
570 		struct inet6_skb_parm	h6;
571 #endif
572 	} header;	/* For incoming frames		*/
573 	__u32		seq;		/* Starting sequence number	*/
574 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
575 	__u32		when;		/* used to compute rtt's	*/
576 	__u8		flags;		/* TCP header flags.		*/
577 
578 	/* NOTE: These must match up to the flags byte in a
579 	 *       real TCP header.
580 	 */
581 #define TCPCB_FLAG_FIN		0x01
582 #define TCPCB_FLAG_SYN		0x02
583 #define TCPCB_FLAG_RST		0x04
584 #define TCPCB_FLAG_PSH		0x08
585 #define TCPCB_FLAG_ACK		0x10
586 #define TCPCB_FLAG_URG		0x20
587 #define TCPCB_FLAG_ECE		0x40
588 #define TCPCB_FLAG_CWR		0x80
589 
590 	__u8		sacked;		/* State flags for SACK/FACK.	*/
591 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
592 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
593 #define TCPCB_LOST		0x04	/* SKB is lost			*/
594 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
595 
596 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
597 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
598 
599 #define TCPCB_URG		0x20	/* Urgent pointer advenced here	*/
600 
601 #define TCPCB_AT_TAIL		(TCPCB_URG)
602 
603 	__u16		urg_ptr;	/* Valid w/URG flags is set.	*/
604 	__u32		ack_seq;	/* Sequence number ACK'd	*/
605 };
606 
607 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
608 
609 #include <net/tcp_ecn.h>
610 
611 /* Due to TSO, an SKB can be composed of multiple actual
612  * packets.  To keep these tracked properly, we use this.
613  */
614 static inline int tcp_skb_pcount(const struct sk_buff *skb)
615 {
616 	return skb_shinfo(skb)->tso_segs;
617 }
618 
619 /* This is valid iff tcp_skb_pcount() > 1. */
620 static inline int tcp_skb_mss(const struct sk_buff *skb)
621 {
622 	return skb_shinfo(skb)->tso_size;
623 }
624 
625 static inline void tcp_dec_pcount_approx(__u32 *count,
626 					 const struct sk_buff *skb)
627 {
628 	if (*count) {
629 		*count -= tcp_skb_pcount(skb);
630 		if ((int)*count < 0)
631 			*count = 0;
632 	}
633 }
634 
635 static inline void tcp_packets_out_inc(struct sock *sk,
636 				       struct tcp_sock *tp,
637 				       const struct sk_buff *skb)
638 {
639 	int orig = tp->packets_out;
640 
641 	tp->packets_out += tcp_skb_pcount(skb);
642 	if (!orig)
643 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
644 					  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
645 }
646 
647 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
648 				       const struct sk_buff *skb)
649 {
650 	tp->packets_out -= tcp_skb_pcount(skb);
651 }
652 
653 /* Events passed to congestion control interface */
654 enum tcp_ca_event {
655 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
656 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
657 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
658 	CA_EVENT_FRTO,		/* fast recovery timeout */
659 	CA_EVENT_LOSS,		/* loss timeout */
660 	CA_EVENT_FAST_ACK,	/* in sequence ack */
661 	CA_EVENT_SLOW_ACK,	/* other ack */
662 };
663 
664 /*
665  * Interface for adding new TCP congestion control handlers
666  */
667 #define TCP_CA_NAME_MAX	16
668 struct tcp_congestion_ops {
669 	struct list_head	list;
670 
671 	/* initialize private data (optional) */
672 	void (*init)(struct sock *sk);
673 	/* cleanup private data  (optional) */
674 	void (*release)(struct sock *sk);
675 
676 	/* return slow start threshold (required) */
677 	u32 (*ssthresh)(struct sock *sk);
678 	/* lower bound for congestion window (optional) */
679 	u32 (*min_cwnd)(struct sock *sk);
680 	/* do new cwnd calculation (required) */
681 	void (*cong_avoid)(struct sock *sk, u32 ack,
682 			   u32 rtt, u32 in_flight, int good_ack);
683 	/* round trip time sample per acked packet (optional) */
684 	void (*rtt_sample)(struct sock *sk, u32 usrtt);
685 	/* call before changing ca_state (optional) */
686 	void (*set_state)(struct sock *sk, u8 new_state);
687 	/* call when cwnd event occurs (optional) */
688 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
689 	/* new value of cwnd after loss (optional) */
690 	u32  (*undo_cwnd)(struct sock *sk);
691 	/* hook for packet ack accounting (optional) */
692 	void (*pkts_acked)(struct sock *sk, u32 num_acked);
693 	/* get info for inet_diag (optional) */
694 	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
695 
696 	char 		name[TCP_CA_NAME_MAX];
697 	struct module 	*owner;
698 };
699 
700 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
701 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
702 
703 extern void tcp_init_congestion_control(struct sock *sk);
704 extern void tcp_cleanup_congestion_control(struct sock *sk);
705 extern int tcp_set_default_congestion_control(const char *name);
706 extern void tcp_get_default_congestion_control(char *name);
707 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
708 
709 extern struct tcp_congestion_ops tcp_init_congestion_ops;
710 extern u32 tcp_reno_ssthresh(struct sock *sk);
711 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
712 				u32 rtt, u32 in_flight, int flag);
713 extern u32 tcp_reno_min_cwnd(struct sock *sk);
714 extern struct tcp_congestion_ops tcp_reno;
715 
716 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
717 {
718 	struct inet_connection_sock *icsk = inet_csk(sk);
719 
720 	if (icsk->icsk_ca_ops->set_state)
721 		icsk->icsk_ca_ops->set_state(sk, ca_state);
722 	icsk->icsk_ca_state = ca_state;
723 }
724 
725 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
726 {
727 	const struct inet_connection_sock *icsk = inet_csk(sk);
728 
729 	if (icsk->icsk_ca_ops->cwnd_event)
730 		icsk->icsk_ca_ops->cwnd_event(sk, event);
731 }
732 
733 /* This determines how many packets are "in the network" to the best
734  * of our knowledge.  In many cases it is conservative, but where
735  * detailed information is available from the receiver (via SACK
736  * blocks etc.) we can make more aggressive calculations.
737  *
738  * Use this for decisions involving congestion control, use just
739  * tp->packets_out to determine if the send queue is empty or not.
740  *
741  * Read this equation as:
742  *
743  *	"Packets sent once on transmission queue" MINUS
744  *	"Packets left network, but not honestly ACKed yet" PLUS
745  *	"Packets fast retransmitted"
746  */
747 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
748 {
749 	return (tp->packets_out - tp->left_out + tp->retrans_out);
750 }
751 
752 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
753  * The exception is rate halving phase, when cwnd is decreasing towards
754  * ssthresh.
755  */
756 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
757 {
758 	const struct tcp_sock *tp = tcp_sk(sk);
759 	if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
760 		return tp->snd_ssthresh;
761 	else
762 		return max(tp->snd_ssthresh,
763 			   ((tp->snd_cwnd >> 1) +
764 			    (tp->snd_cwnd >> 2)));
765 }
766 
767 static inline void tcp_sync_left_out(struct tcp_sock *tp)
768 {
769 	if (tp->rx_opt.sack_ok &&
770 	    (tp->sacked_out >= tp->packets_out - tp->lost_out))
771 		tp->sacked_out = tp->packets_out - tp->lost_out;
772 	tp->left_out = tp->sacked_out + tp->lost_out;
773 }
774 
775 /* Set slow start threshold and cwnd not falling to slow start */
776 static inline void __tcp_enter_cwr(struct sock *sk)
777 {
778 	const struct inet_connection_sock *icsk = inet_csk(sk);
779 	struct tcp_sock *tp = tcp_sk(sk);
780 
781 	tp->undo_marker = 0;
782 	tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
783 	tp->snd_cwnd = min(tp->snd_cwnd,
784 			   tcp_packets_in_flight(tp) + 1U);
785 	tp->snd_cwnd_cnt = 0;
786 	tp->high_seq = tp->snd_nxt;
787 	tp->snd_cwnd_stamp = tcp_time_stamp;
788 	TCP_ECN_queue_cwr(tp);
789 }
790 
791 static inline void tcp_enter_cwr(struct sock *sk)
792 {
793 	struct tcp_sock *tp = tcp_sk(sk);
794 
795 	tp->prior_ssthresh = 0;
796 	if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {
797 		__tcp_enter_cwr(sk);
798 		tcp_set_ca_state(sk, TCP_CA_CWR);
799 	}
800 }
801 
802 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
803 
804 /* Slow start with delack produces 3 packets of burst, so that
805  * it is safe "de facto".
806  */
807 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
808 {
809 	return 3;
810 }
811 
812 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
813 					   const struct sk_buff *skb)
814 {
815 	if (skb->len < mss)
816 		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
817 }
818 
819 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
820 {
821 	const struct inet_connection_sock *icsk = inet_csk(sk);
822 	if (!tp->packets_out && !icsk->icsk_pending)
823 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
824 					  icsk->icsk_rto, TCP_RTO_MAX);
825 }
826 
827 static __inline__ void tcp_push_pending_frames(struct sock *sk,
828 					       struct tcp_sock *tp)
829 {
830 	__tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
831 }
832 
833 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
834 {
835 	tp->snd_wl1 = seq;
836 }
837 
838 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
839 {
840 	tp->snd_wl1 = seq;
841 }
842 
843 /*
844  * Calculate(/check) TCP checksum
845  */
846 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
847 				   unsigned long saddr, unsigned long daddr,
848 				   unsigned long base)
849 {
850 	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
851 }
852 
853 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
854 {
855 	return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
856 }
857 
858 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
859 {
860 	return skb->ip_summed != CHECKSUM_UNNECESSARY &&
861 		__tcp_checksum_complete(skb);
862 }
863 
864 /* Prequeue for VJ style copy to user, combined with checksumming. */
865 
866 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
867 {
868 	tp->ucopy.task = NULL;
869 	tp->ucopy.len = 0;
870 	tp->ucopy.memory = 0;
871 	skb_queue_head_init(&tp->ucopy.prequeue);
872 }
873 
874 /* Packet is added to VJ-style prequeue for processing in process
875  * context, if a reader task is waiting. Apparently, this exciting
876  * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
877  * failed somewhere. Latency? Burstiness? Well, at least now we will
878  * see, why it failed. 8)8)				  --ANK
879  *
880  * NOTE: is this not too big to inline?
881  */
882 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
883 {
884 	struct tcp_sock *tp = tcp_sk(sk);
885 
886 	if (!sysctl_tcp_low_latency && tp->ucopy.task) {
887 		__skb_queue_tail(&tp->ucopy.prequeue, skb);
888 		tp->ucopy.memory += skb->truesize;
889 		if (tp->ucopy.memory > sk->sk_rcvbuf) {
890 			struct sk_buff *skb1;
891 
892 			BUG_ON(sock_owned_by_user(sk));
893 
894 			while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
895 				sk->sk_backlog_rcv(sk, skb1);
896 				NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
897 			}
898 
899 			tp->ucopy.memory = 0;
900 		} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
901 			wake_up_interruptible(sk->sk_sleep);
902 			if (!inet_csk_ack_scheduled(sk))
903 				inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
904 						          (3 * TCP_RTO_MIN) / 4,
905 							  TCP_RTO_MAX);
906 		}
907 		return 1;
908 	}
909 	return 0;
910 }
911 
912 
913 #undef STATE_TRACE
914 
915 #ifdef STATE_TRACE
916 static const char *statename[]={
917 	"Unused","Established","Syn Sent","Syn Recv",
918 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
919 	"Close Wait","Last ACK","Listen","Closing"
920 };
921 #endif
922 
923 static __inline__ void tcp_set_state(struct sock *sk, int state)
924 {
925 	int oldstate = sk->sk_state;
926 
927 	switch (state) {
928 	case TCP_ESTABLISHED:
929 		if (oldstate != TCP_ESTABLISHED)
930 			TCP_INC_STATS(TCP_MIB_CURRESTAB);
931 		break;
932 
933 	case TCP_CLOSE:
934 		if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
935 			TCP_INC_STATS(TCP_MIB_ESTABRESETS);
936 
937 		sk->sk_prot->unhash(sk);
938 		if (inet_csk(sk)->icsk_bind_hash &&
939 		    !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
940 			inet_put_port(&tcp_hashinfo, sk);
941 		/* fall through */
942 	default:
943 		if (oldstate==TCP_ESTABLISHED)
944 			TCP_DEC_STATS(TCP_MIB_CURRESTAB);
945 	}
946 
947 	/* Change state AFTER socket is unhashed to avoid closed
948 	 * socket sitting in hash tables.
949 	 */
950 	sk->sk_state = state;
951 
952 #ifdef STATE_TRACE
953 	SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
954 #endif
955 }
956 
957 static __inline__ void tcp_done(struct sock *sk)
958 {
959 	tcp_set_state(sk, TCP_CLOSE);
960 	tcp_clear_xmit_timers(sk);
961 
962 	sk->sk_shutdown = SHUTDOWN_MASK;
963 
964 	if (!sock_flag(sk, SOCK_DEAD))
965 		sk->sk_state_change(sk);
966 	else
967 		inet_csk_destroy_sock(sk);
968 }
969 
970 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
971 {
972 	rx_opt->dsack = 0;
973 	rx_opt->eff_sacks = 0;
974 	rx_opt->num_sacks = 0;
975 }
976 
977 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
978 {
979 	if (tp->rx_opt.tstamp_ok) {
980 		*ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
981 					  (TCPOPT_NOP << 16) |
982 					  (TCPOPT_TIMESTAMP << 8) |
983 					  TCPOLEN_TIMESTAMP);
984 		*ptr++ = htonl(tstamp);
985 		*ptr++ = htonl(tp->rx_opt.ts_recent);
986 	}
987 	if (tp->rx_opt.eff_sacks) {
988 		struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
989 		int this_sack;
990 
991 		*ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
992 					  (TCPOPT_NOP << 16) |
993 					  (TCPOPT_SACK << 8) |
994 					  (TCPOLEN_SACK_BASE +
995 					   (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
996 		for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
997 			*ptr++ = htonl(sp[this_sack].start_seq);
998 			*ptr++ = htonl(sp[this_sack].end_seq);
999 		}
1000 		if (tp->rx_opt.dsack) {
1001 			tp->rx_opt.dsack = 0;
1002 			tp->rx_opt.eff_sacks--;
1003 		}
1004 	}
1005 }
1006 
1007 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1008  * If this is every changed make sure to change the definition of
1009  * MAX_SYN_SIZE to match the new maximum number of options that you
1010  * can generate.
1011  */
1012 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1013 					     int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1014 {
1015 	/* We always get an MSS option.
1016 	 * The option bytes which will be seen in normal data
1017 	 * packets should timestamps be used, must be in the MSS
1018 	 * advertised.  But we subtract them from tp->mss_cache so
1019 	 * that calculations in tcp_sendmsg are simpler etc.
1020 	 * So account for this fact here if necessary.  If we
1021 	 * don't do this correctly, as a receiver we won't
1022 	 * recognize data packets as being full sized when we
1023 	 * should, and thus we won't abide by the delayed ACK
1024 	 * rules correctly.
1025 	 * SACKs don't matter, we never delay an ACK when we
1026 	 * have any of those going out.
1027 	 */
1028 	*ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1029 	if (ts) {
1030 		if(sack)
1031 			*ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1032 						  (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1033 		else
1034 			*ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1035 						  (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1036 		*ptr++ = htonl(tstamp);		/* TSVAL */
1037 		*ptr++ = htonl(ts_recent);	/* TSECR */
1038 	} else if(sack)
1039 		*ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1040 					  (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1041 	if (offer_wscale)
1042 		*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1043 }
1044 
1045 /* Determine a window scaling and initial window to offer. */
1046 extern void tcp_select_initial_window(int __space, __u32 mss,
1047 				      __u32 *rcv_wnd, __u32 *window_clamp,
1048 				      int wscale_ok, __u8 *rcv_wscale);
1049 
1050 static inline int tcp_win_from_space(int space)
1051 {
1052 	return sysctl_tcp_adv_win_scale<=0 ?
1053 		(space>>(-sysctl_tcp_adv_win_scale)) :
1054 		space - (space>>sysctl_tcp_adv_win_scale);
1055 }
1056 
1057 /* Note: caller must be prepared to deal with negative returns */
1058 static inline int tcp_space(const struct sock *sk)
1059 {
1060 	return tcp_win_from_space(sk->sk_rcvbuf -
1061 				  atomic_read(&sk->sk_rmem_alloc));
1062 }
1063 
1064 static inline int tcp_full_space(const struct sock *sk)
1065 {
1066 	return tcp_win_from_space(sk->sk_rcvbuf);
1067 }
1068 
1069 static __inline__ void tcp_openreq_init(struct request_sock *req,
1070 					struct tcp_options_received *rx_opt,
1071 					struct sk_buff *skb)
1072 {
1073 	struct inet_request_sock *ireq = inet_rsk(req);
1074 
1075 	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
1076 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1077 	req->mss = rx_opt->mss_clamp;
1078 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1079 	ireq->tstamp_ok = rx_opt->tstamp_ok;
1080 	ireq->sack_ok = rx_opt->sack_ok;
1081 	ireq->snd_wscale = rx_opt->snd_wscale;
1082 	ireq->wscale_ok = rx_opt->wscale_ok;
1083 	ireq->acked = 0;
1084 	ireq->ecn_ok = 0;
1085 	ireq->rmt_port = skb->h.th->source;
1086 }
1087 
1088 extern void tcp_enter_memory_pressure(void);
1089 
1090 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1091 {
1092 	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1093 }
1094 
1095 static inline int keepalive_time_when(const struct tcp_sock *tp)
1096 {
1097 	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1098 }
1099 
1100 static inline int tcp_fin_time(const struct sock *sk)
1101 {
1102 	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1103 	const int rto = inet_csk(sk)->icsk_rto;
1104 
1105 	if (fin_timeout < (rto << 2) - (rto >> 1))
1106 		fin_timeout = (rto << 2) - (rto >> 1);
1107 
1108 	return fin_timeout;
1109 }
1110 
1111 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1112 {
1113 	if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1114 		return 0;
1115 	if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1116 		return 0;
1117 
1118 	/* RST segments are not recommended to carry timestamp,
1119 	   and, if they do, it is recommended to ignore PAWS because
1120 	   "their cleanup function should take precedence over timestamps."
1121 	   Certainly, it is mistake. It is necessary to understand the reasons
1122 	   of this constraint to relax it: if peer reboots, clock may go
1123 	   out-of-sync and half-open connections will not be reset.
1124 	   Actually, the problem would be not existing if all
1125 	   the implementations followed draft about maintaining clock
1126 	   via reboots. Linux-2.2 DOES NOT!
1127 
1128 	   However, we can relax time bounds for RST segments to MSL.
1129 	 */
1130 	if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1131 		return 0;
1132 	return 1;
1133 }
1134 
1135 #define TCP_CHECK_TIMER(sk) do { } while (0)
1136 
1137 static inline int tcp_use_frto(const struct sock *sk)
1138 {
1139 	const struct tcp_sock *tp = tcp_sk(sk);
1140 
1141 	/* F-RTO must be activated in sysctl and there must be some
1142 	 * unsent new data, and the advertised window should allow
1143 	 * sending it.
1144 	 */
1145 	return (sysctl_tcp_frto && sk->sk_send_head &&
1146 		!after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1147 		       tp->snd_una + tp->snd_wnd));
1148 }
1149 
1150 static inline void tcp_mib_init(void)
1151 {
1152 	/* See RFC 2012 */
1153 	TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1154 	TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1155 	TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1156 	TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1157 }
1158 
1159 /* /proc */
1160 enum tcp_seq_states {
1161 	TCP_SEQ_STATE_LISTENING,
1162 	TCP_SEQ_STATE_OPENREQ,
1163 	TCP_SEQ_STATE_ESTABLISHED,
1164 	TCP_SEQ_STATE_TIME_WAIT,
1165 };
1166 
1167 struct tcp_seq_afinfo {
1168 	struct module		*owner;
1169 	char			*name;
1170 	sa_family_t		family;
1171 	int			(*seq_show) (struct seq_file *m, void *v);
1172 	struct file_operations	*seq_fops;
1173 };
1174 
1175 struct tcp_iter_state {
1176 	sa_family_t		family;
1177 	enum tcp_seq_states	state;
1178 	struct sock		*syn_wait_sk;
1179 	int			bucket, sbucket, num, uid;
1180 	struct seq_operations	seq_ops;
1181 };
1182 
1183 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1184 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1185 
1186 extern struct request_sock_ops tcp_request_sock_ops;
1187 
1188 extern int tcp_v4_destroy_sock(struct sock *sk);
1189 
1190 #ifdef CONFIG_PROC_FS
1191 extern int  tcp4_proc_init(void);
1192 extern void tcp4_proc_exit(void);
1193 #endif
1194 
1195 extern void tcp_v4_init(struct net_proto_family *ops);
1196 extern void tcp_init(void);
1197 
1198 #endif	/* _TCP_H */
1199