xref: /linux/include/net/tcp.h (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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/list.h>
25 #include <linux/tcp.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 
33 #include <net/inet_connection_sock.h>
34 #include <net/inet_timewait_sock.h>
35 #include <net/inet_hashtables.h>
36 #include <net/checksum.h>
37 #include <net/request_sock.h>
38 #include <net/sock.h>
39 #include <net/snmp.h>
40 #include <net/ip.h>
41 #include <net/tcp_states.h>
42 
43 #include <linux/seq_file.h>
44 
45 extern struct inet_hashinfo tcp_hashinfo;
46 
47 extern atomic_t tcp_orphan_count;
48 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
49 
50 #define MAX_TCP_HEADER	(128 + MAX_HEADER)
51 
52 /*
53  * Never offer a window over 32767 without using window scaling. Some
54  * poor stacks do signed 16bit maths!
55  */
56 #define MAX_TCP_WINDOW		32767U
57 
58 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59 #define TCP_MIN_MSS		88U
60 
61 /* Minimal RCV_MSS. */
62 #define TCP_MIN_RCVMSS		536U
63 
64 /* The least MTU to use for probing */
65 #define TCP_BASE_MSS		512
66 
67 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
68 #define TCP_FASTRETRANS_THRESH 3
69 
70 /* Maximal reordering. */
71 #define TCP_MAX_REORDERING	127
72 
73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
74 #define TCP_MAX_QUICKACKS	16U
75 
76 /* urg_data states */
77 #define TCP_URG_VALID	0x0100
78 #define TCP_URG_NOTYET	0x0200
79 #define TCP_URG_READ	0x0400
80 
81 #define TCP_RETR1	3	/*
82 				 * This is how many retries it does before it
83 				 * tries to figure out if the gateway is
84 				 * down. Minimal RFC value is 3; it corresponds
85 				 * to ~3sec-8min depending on RTO.
86 				 */
87 
88 #define TCP_RETR2	15	/*
89 				 * This should take at least
90 				 * 90 minutes to time out.
91 				 * RFC1122 says that the limit is 100 sec.
92 				 * 15 is ~13-30min depending on RTO.
93 				 */
94 
95 #define TCP_SYN_RETRIES	 5	/* number of times to retry active opening a
96 				 * connection: ~180sec is RFC minimum	*/
97 
98 #define TCP_SYNACK_RETRIES 5	/* number of times to retry passive opening a
99 				 * connection: ~180sec is RFC minimum	*/
100 
101 
102 #define TCP_ORPHAN_RETRIES 7	/* number of times to retry on an orphaned
103 				 * socket. 7 is ~50sec-16min.
104 				 */
105 
106 
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 				  * state, about 60 seconds	*/
109 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
110                                  /* BSD style FIN_WAIT2 deadlock breaker.
111 				  * It used to be 3min, new value is 60sec,
112 				  * to combine FIN-WAIT-2 timeout with
113 				  * TIME-WAIT timer.
114 				  */
115 
116 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
117 #if HZ >= 100
118 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN	((unsigned)(HZ/25))
120 #else
121 #define TCP_DELACK_MIN	4U
122 #define TCP_ATO_MIN	4U
123 #endif
124 #define TCP_RTO_MAX	((unsigned)(120*HZ))
125 #define TCP_RTO_MIN	((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value	*/
127 
128 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
129 					                 * for local resources.
130 					                 */
131 
132 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
133 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
134 #define TCP_KEEPALIVE_INTVL	(75*HZ)
135 
136 #define MAX_TCP_KEEPIDLE	32767
137 #define MAX_TCP_KEEPINTVL	32767
138 #define MAX_TCP_KEEPCNT		127
139 #define MAX_TCP_SYNCNT		127
140 
141 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
142 
143 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
144 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
145 					 * after this time. It should be equal
146 					 * (or greater than) TCP_TIMEWAIT_LEN
147 					 * to provide reliability equal to one
148 					 * provided by timewait state.
149 					 */
150 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
151 					 * timestamps. It must be less than
152 					 * minimal timewait lifetime.
153 					 */
154 /*
155  *	TCP option
156  */
157 
158 #define TCPOPT_NOP		1	/* Padding */
159 #define TCPOPT_EOL		0	/* End of options */
160 #define TCPOPT_MSS		2	/* Segment size negotiating */
161 #define TCPOPT_WINDOW		3	/* Window scaling */
162 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
163 #define TCPOPT_SACK             5       /* SACK Block */
164 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
165 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
166 
167 /*
168  *     TCP option lengths
169  */
170 
171 #define TCPOLEN_MSS            4
172 #define TCPOLEN_WINDOW         3
173 #define TCPOLEN_SACK_PERM      2
174 #define TCPOLEN_TIMESTAMP      10
175 #define TCPOLEN_MD5SIG         18
176 
177 /* But this is what stacks really send out. */
178 #define TCPOLEN_TSTAMP_ALIGNED		12
179 #define TCPOLEN_WSCALE_ALIGNED		4
180 #define TCPOLEN_SACKPERM_ALIGNED	4
181 #define TCPOLEN_SACK_BASE		2
182 #define TCPOLEN_SACK_BASE_ALIGNED	4
183 #define TCPOLEN_SACK_PERBLOCK		8
184 #define TCPOLEN_MD5SIG_ALIGNED		20
185 
186 /* Flags in tp->nonagle */
187 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
188 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
189 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
190 
191 extern struct inet_timewait_death_row tcp_death_row;
192 
193 /* sysctl variables for tcp */
194 extern int sysctl_tcp_timestamps;
195 extern int sysctl_tcp_window_scaling;
196 extern int sysctl_tcp_sack;
197 extern int sysctl_tcp_fin_timeout;
198 extern int sysctl_tcp_keepalive_time;
199 extern int sysctl_tcp_keepalive_probes;
200 extern int sysctl_tcp_keepalive_intvl;
201 extern int sysctl_tcp_syn_retries;
202 extern int sysctl_tcp_synack_retries;
203 extern int sysctl_tcp_retries1;
204 extern int sysctl_tcp_retries2;
205 extern int sysctl_tcp_orphan_retries;
206 extern int sysctl_tcp_syncookies;
207 extern int sysctl_tcp_retrans_collapse;
208 extern int sysctl_tcp_stdurg;
209 extern int sysctl_tcp_rfc1337;
210 extern int sysctl_tcp_abort_on_overflow;
211 extern int sysctl_tcp_max_orphans;
212 extern int sysctl_tcp_fack;
213 extern int sysctl_tcp_reordering;
214 extern int sysctl_tcp_ecn;
215 extern int sysctl_tcp_dsack;
216 extern int sysctl_tcp_mem[3];
217 extern int sysctl_tcp_wmem[3];
218 extern int sysctl_tcp_rmem[3];
219 extern int sysctl_tcp_app_win;
220 extern int sysctl_tcp_adv_win_scale;
221 extern int sysctl_tcp_tw_reuse;
222 extern int sysctl_tcp_frto;
223 extern int sysctl_tcp_low_latency;
224 extern int sysctl_tcp_dma_copybreak;
225 extern int sysctl_tcp_nometrics_save;
226 extern int sysctl_tcp_moderate_rcvbuf;
227 extern int sysctl_tcp_tso_win_divisor;
228 extern int sysctl_tcp_abc;
229 extern int sysctl_tcp_mtu_probing;
230 extern int sysctl_tcp_base_mss;
231 extern int sysctl_tcp_workaround_signed_windows;
232 extern int sysctl_tcp_slow_start_after_idle;
233 
234 extern atomic_t tcp_memory_allocated;
235 extern atomic_t tcp_sockets_allocated;
236 extern int tcp_memory_pressure;
237 
238 /*
239  * The next routines deal with comparing 32 bit unsigned ints
240  * and worry about wraparound (automatic with unsigned arithmetic).
241  */
242 
243 static inline int before(__u32 seq1, __u32 seq2)
244 {
245         return (__s32)(seq1-seq2) < 0;
246 }
247 #define after(seq2, seq1) 	before(seq1, seq2)
248 
249 /* is s2<=s1<=s3 ? */
250 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
251 {
252 	return seq3 - seq2 >= seq1 - seq2;
253 }
254 
255 
256 extern struct proto tcp_prot;
257 
258 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
259 #define TCP_INC_STATS(field)		SNMP_INC_STATS(tcp_statistics, field)
260 #define TCP_INC_STATS_BH(field)		SNMP_INC_STATS_BH(tcp_statistics, field)
261 #define TCP_INC_STATS_USER(field) 	SNMP_INC_STATS_USER(tcp_statistics, field)
262 #define TCP_DEC_STATS(field)		SNMP_DEC_STATS(tcp_statistics, field)
263 #define TCP_ADD_STATS_BH(field, val)	SNMP_ADD_STATS_BH(tcp_statistics, field, val)
264 #define TCP_ADD_STATS_USER(field, val)	SNMP_ADD_STATS_USER(tcp_statistics, field, val)
265 
266 extern void			tcp_v4_err(struct sk_buff *skb, u32);
267 
268 extern void			tcp_shutdown (struct sock *sk, int how);
269 
270 extern int			tcp_v4_rcv(struct sk_buff *skb);
271 
272 extern int			tcp_v4_remember_stamp(struct sock *sk);
273 
274 extern int		    	tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
275 
276 extern int			tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
277 					    struct msghdr *msg, size_t size);
278 extern ssize_t			tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
279 
280 extern int			tcp_ioctl(struct sock *sk,
281 					  int cmd,
282 					  unsigned long arg);
283 
284 extern int			tcp_rcv_state_process(struct sock *sk,
285 						      struct sk_buff *skb,
286 						      struct tcphdr *th,
287 						      unsigned len);
288 
289 extern int			tcp_rcv_established(struct sock *sk,
290 						    struct sk_buff *skb,
291 						    struct tcphdr *th,
292 						    unsigned len);
293 
294 extern void			tcp_rcv_space_adjust(struct sock *sk);
295 
296 extern void			tcp_cleanup_rbuf(struct sock *sk, int copied);
297 
298 extern int			tcp_twsk_unique(struct sock *sk,
299 						struct sock *sktw, void *twp);
300 
301 extern void			tcp_twsk_destructor(struct sock *sk);
302 
303 static inline void tcp_dec_quickack_mode(struct sock *sk,
304 					 const unsigned int pkts)
305 {
306 	struct inet_connection_sock *icsk = inet_csk(sk);
307 
308 	if (icsk->icsk_ack.quick) {
309 		if (pkts >= icsk->icsk_ack.quick) {
310 			icsk->icsk_ack.quick = 0;
311 			/* Leaving quickack mode we deflate ATO. */
312 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
313 		} else
314 			icsk->icsk_ack.quick -= pkts;
315 	}
316 }
317 
318 extern void tcp_enter_quickack_mode(struct sock *sk);
319 
320 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
321 {
322  	rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
323 }
324 
325 enum tcp_tw_status
326 {
327 	TCP_TW_SUCCESS = 0,
328 	TCP_TW_RST = 1,
329 	TCP_TW_ACK = 2,
330 	TCP_TW_SYN = 3
331 };
332 
333 
334 extern enum tcp_tw_status	tcp_timewait_state_process(struct inet_timewait_sock *tw,
335 							   struct sk_buff *skb,
336 							   const struct tcphdr *th);
337 
338 extern struct sock *		tcp_check_req(struct sock *sk,struct sk_buff *skb,
339 					      struct request_sock *req,
340 					      struct request_sock **prev);
341 extern int			tcp_child_process(struct sock *parent,
342 						  struct sock *child,
343 						  struct sk_buff *skb);
344 extern void			tcp_enter_frto(struct sock *sk);
345 extern void			tcp_enter_loss(struct sock *sk, int how);
346 extern void			tcp_clear_retrans(struct tcp_sock *tp);
347 extern void			tcp_update_metrics(struct sock *sk);
348 
349 extern void			tcp_close(struct sock *sk,
350 					  long timeout);
351 extern unsigned int		tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
352 
353 extern int			tcp_getsockopt(struct sock *sk, int level,
354 					       int optname,
355 					       char __user *optval,
356 					       int __user *optlen);
357 extern int			tcp_setsockopt(struct sock *sk, int level,
358 					       int optname, char __user *optval,
359 					       int optlen);
360 extern int			compat_tcp_getsockopt(struct sock *sk,
361 					int level, int optname,
362 					char __user *optval, int __user *optlen);
363 extern int			compat_tcp_setsockopt(struct sock *sk,
364 					int level, int optname,
365 					char __user *optval, int optlen);
366 extern void			tcp_set_keepalive(struct sock *sk, int val);
367 extern int			tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
368 					    struct msghdr *msg,
369 					    size_t len, int nonblock,
370 					    int flags, int *addr_len);
371 
372 extern void			tcp_parse_options(struct sk_buff *skb,
373 						  struct tcp_options_received *opt_rx,
374 						  int estab);
375 
376 /*
377  *	TCP v4 functions exported for the inet6 API
378  */
379 
380 extern void		       	tcp_v4_send_check(struct sock *sk, int len,
381 						  struct sk_buff *skb);
382 
383 extern int			tcp_v4_conn_request(struct sock *sk,
384 						    struct sk_buff *skb);
385 
386 extern struct sock *		tcp_create_openreq_child(struct sock *sk,
387 							 struct request_sock *req,
388 							 struct sk_buff *skb);
389 
390 extern struct sock *		tcp_v4_syn_recv_sock(struct sock *sk,
391 						     struct sk_buff *skb,
392 						     struct request_sock *req,
393 							struct dst_entry *dst);
394 
395 extern int			tcp_v4_do_rcv(struct sock *sk,
396 					      struct sk_buff *skb);
397 
398 extern int			tcp_v4_connect(struct sock *sk,
399 					       struct sockaddr *uaddr,
400 					       int addr_len);
401 
402 extern int			tcp_connect(struct sock *sk);
403 
404 extern struct sk_buff *		tcp_make_synack(struct sock *sk,
405 						struct dst_entry *dst,
406 						struct request_sock *req);
407 
408 extern int			tcp_disconnect(struct sock *sk, int flags);
409 
410 extern void			tcp_unhash(struct sock *sk);
411 
412 /* From syncookies.c */
413 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
414 				    struct ip_options *opt);
415 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
416 				     __u16 *mss);
417 
418 /* tcp_output.c */
419 
420 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
421 				      unsigned int cur_mss, int nonagle);
422 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
423 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
424 extern void tcp_xmit_retransmit_queue(struct sock *);
425 extern void tcp_simple_retransmit(struct sock *);
426 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
427 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
428 
429 extern void tcp_send_probe0(struct sock *);
430 extern void tcp_send_partial(struct sock *);
431 extern int  tcp_write_wakeup(struct sock *);
432 extern void tcp_send_fin(struct sock *sk);
433 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
434 extern int  tcp_send_synack(struct sock *);
435 extern void tcp_push_one(struct sock *, unsigned int mss_now);
436 extern void tcp_send_ack(struct sock *sk);
437 extern void tcp_send_delayed_ack(struct sock *sk);
438 
439 /* tcp_input.c */
440 extern void tcp_cwnd_application_limited(struct sock *sk);
441 
442 /* tcp_timer.c */
443 extern void tcp_init_xmit_timers(struct sock *);
444 static inline void tcp_clear_xmit_timers(struct sock *sk)
445 {
446 	inet_csk_clear_xmit_timers(sk);
447 }
448 
449 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
450 extern unsigned int tcp_current_mss(struct sock *sk, int large);
451 
452 /* tcp.c */
453 extern void tcp_get_info(struct sock *, struct tcp_info *);
454 
455 /* Read 'sendfile()'-style from a TCP socket */
456 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
457 				unsigned int, size_t);
458 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
459 			 sk_read_actor_t recv_actor);
460 
461 extern void tcp_initialize_rcv_mss(struct sock *sk);
462 
463 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
464 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
465 extern void tcp_mtup_init(struct sock *sk);
466 
467 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
468 {
469 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
470 			       ntohl(TCP_FLAG_ACK) |
471 			       snd_wnd);
472 }
473 
474 static inline void tcp_fast_path_on(struct tcp_sock *tp)
475 {
476 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
477 }
478 
479 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
480 {
481 	if (skb_queue_empty(&tp->out_of_order_queue) &&
482 	    tp->rcv_wnd &&
483 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
484 	    !tp->urg_data)
485 		tcp_fast_path_on(tp);
486 }
487 
488 /* Compute the actual receive window we are currently advertising.
489  * Rcv_nxt can be after the window if our peer push more data
490  * than the offered window.
491  */
492 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
493 {
494 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
495 
496 	if (win < 0)
497 		win = 0;
498 	return (u32) win;
499 }
500 
501 /* Choose a new window, without checks for shrinking, and without
502  * scaling applied to the result.  The caller does these things
503  * if necessary.  This is a "raw" window selection.
504  */
505 extern u32	__tcp_select_window(struct sock *sk);
506 
507 /* TCP timestamps are only 32-bits, this causes a slight
508  * complication on 64-bit systems since we store a snapshot
509  * of jiffies in the buffer control blocks below.  We decided
510  * to use only the low 32-bits of jiffies and hide the ugly
511  * casts with the following macro.
512  */
513 #define tcp_time_stamp		((__u32)(jiffies))
514 
515 /* This is what the send packet queuing engine uses to pass
516  * TCP per-packet control information to the transmission
517  * code.  We also store the host-order sequence numbers in
518  * here too.  This is 36 bytes on 32-bit architectures,
519  * 40 bytes on 64-bit machines, if this grows please adjust
520  * skbuff.h:skbuff->cb[xxx] size appropriately.
521  */
522 struct tcp_skb_cb {
523 	union {
524 		struct inet_skb_parm	h4;
525 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
526 		struct inet6_skb_parm	h6;
527 #endif
528 	} header;	/* For incoming frames		*/
529 	__u32		seq;		/* Starting sequence number	*/
530 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
531 	__u32		when;		/* used to compute rtt's	*/
532 	__u8		flags;		/* TCP header flags.		*/
533 
534 	/* NOTE: These must match up to the flags byte in a
535 	 *       real TCP header.
536 	 */
537 #define TCPCB_FLAG_FIN		0x01
538 #define TCPCB_FLAG_SYN		0x02
539 #define TCPCB_FLAG_RST		0x04
540 #define TCPCB_FLAG_PSH		0x08
541 #define TCPCB_FLAG_ACK		0x10
542 #define TCPCB_FLAG_URG		0x20
543 #define TCPCB_FLAG_ECE		0x40
544 #define TCPCB_FLAG_CWR		0x80
545 
546 	__u8		sacked;		/* State flags for SACK/FACK.	*/
547 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
548 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
549 #define TCPCB_LOST		0x04	/* SKB is lost			*/
550 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
551 
552 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
553 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
554 
555 #define TCPCB_URG		0x20	/* Urgent pointer advanced here	*/
556 
557 #define TCPCB_AT_TAIL		(TCPCB_URG)
558 
559 	__u16		urg_ptr;	/* Valid w/URG flags is set.	*/
560 	__u32		ack_seq;	/* Sequence number ACK'd	*/
561 };
562 
563 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
564 
565 #include <net/tcp_ecn.h>
566 
567 /* Due to TSO, an SKB can be composed of multiple actual
568  * packets.  To keep these tracked properly, we use this.
569  */
570 static inline int tcp_skb_pcount(const struct sk_buff *skb)
571 {
572 	return skb_shinfo(skb)->gso_segs;
573 }
574 
575 /* This is valid iff tcp_skb_pcount() > 1. */
576 static inline int tcp_skb_mss(const struct sk_buff *skb)
577 {
578 	return skb_shinfo(skb)->gso_size;
579 }
580 
581 static inline void tcp_dec_pcount_approx(__u32 *count,
582 					 const struct sk_buff *skb)
583 {
584 	if (*count) {
585 		*count -= tcp_skb_pcount(skb);
586 		if ((int)*count < 0)
587 			*count = 0;
588 	}
589 }
590 
591 static inline void tcp_packets_out_inc(struct sock *sk,
592 				       struct tcp_sock *tp,
593 				       const struct sk_buff *skb)
594 {
595 	int orig = tp->packets_out;
596 
597 	tp->packets_out += tcp_skb_pcount(skb);
598 	if (!orig)
599 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
600 					  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
601 }
602 
603 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
604 				       const struct sk_buff *skb)
605 {
606 	tp->packets_out -= tcp_skb_pcount(skb);
607 }
608 
609 /* Events passed to congestion control interface */
610 enum tcp_ca_event {
611 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
612 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
613 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
614 	CA_EVENT_FRTO,		/* fast recovery timeout */
615 	CA_EVENT_LOSS,		/* loss timeout */
616 	CA_EVENT_FAST_ACK,	/* in sequence ack */
617 	CA_EVENT_SLOW_ACK,	/* other ack */
618 };
619 
620 /*
621  * Interface for adding new TCP congestion control handlers
622  */
623 #define TCP_CA_NAME_MAX	16
624 #define TCP_CA_MAX	128
625 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
626 
627 struct tcp_congestion_ops {
628 	struct list_head	list;
629 	int	non_restricted;
630 
631 	/* initialize private data (optional) */
632 	void (*init)(struct sock *sk);
633 	/* cleanup private data  (optional) */
634 	void (*release)(struct sock *sk);
635 
636 	/* return slow start threshold (required) */
637 	u32 (*ssthresh)(struct sock *sk);
638 	/* lower bound for congestion window (optional) */
639 	u32 (*min_cwnd)(const struct sock *sk);
640 	/* do new cwnd calculation (required) */
641 	void (*cong_avoid)(struct sock *sk, u32 ack,
642 			   u32 rtt, u32 in_flight, int good_ack);
643 	/* round trip time sample per acked packet (optional) */
644 	void (*rtt_sample)(struct sock *sk, u32 usrtt);
645 	/* call before changing ca_state (optional) */
646 	void (*set_state)(struct sock *sk, u8 new_state);
647 	/* call when cwnd event occurs (optional) */
648 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
649 	/* new value of cwnd after loss (optional) */
650 	u32  (*undo_cwnd)(struct sock *sk);
651 	/* hook for packet ack accounting (optional) */
652 	void (*pkts_acked)(struct sock *sk, u32 num_acked);
653 	/* get info for inet_diag (optional) */
654 	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
655 
656 	char 		name[TCP_CA_NAME_MAX];
657 	struct module 	*owner;
658 };
659 
660 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
661 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
662 
663 extern void tcp_init_congestion_control(struct sock *sk);
664 extern void tcp_cleanup_congestion_control(struct sock *sk);
665 extern int tcp_set_default_congestion_control(const char *name);
666 extern void tcp_get_default_congestion_control(char *name);
667 extern void tcp_get_available_congestion_control(char *buf, size_t len);
668 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
669 extern int tcp_set_allowed_congestion_control(char *allowed);
670 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
671 extern void tcp_slow_start(struct tcp_sock *tp);
672 
673 extern struct tcp_congestion_ops tcp_init_congestion_ops;
674 extern u32 tcp_reno_ssthresh(struct sock *sk);
675 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
676 				u32 rtt, u32 in_flight, int flag);
677 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
678 extern struct tcp_congestion_ops tcp_reno;
679 
680 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
681 {
682 	struct inet_connection_sock *icsk = inet_csk(sk);
683 
684 	if (icsk->icsk_ca_ops->set_state)
685 		icsk->icsk_ca_ops->set_state(sk, ca_state);
686 	icsk->icsk_ca_state = ca_state;
687 }
688 
689 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
690 {
691 	const struct inet_connection_sock *icsk = inet_csk(sk);
692 
693 	if (icsk->icsk_ca_ops->cwnd_event)
694 		icsk->icsk_ca_ops->cwnd_event(sk, event);
695 }
696 
697 /* This determines how many packets are "in the network" to the best
698  * of our knowledge.  In many cases it is conservative, but where
699  * detailed information is available from the receiver (via SACK
700  * blocks etc.) we can make more aggressive calculations.
701  *
702  * Use this for decisions involving congestion control, use just
703  * tp->packets_out to determine if the send queue is empty or not.
704  *
705  * Read this equation as:
706  *
707  *	"Packets sent once on transmission queue" MINUS
708  *	"Packets left network, but not honestly ACKed yet" PLUS
709  *	"Packets fast retransmitted"
710  */
711 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
712 {
713 	return (tp->packets_out - tp->left_out + tp->retrans_out);
714 }
715 
716 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
717  * The exception is rate halving phase, when cwnd is decreasing towards
718  * ssthresh.
719  */
720 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
721 {
722 	const struct tcp_sock *tp = tcp_sk(sk);
723 	if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
724 		return tp->snd_ssthresh;
725 	else
726 		return max(tp->snd_ssthresh,
727 			   ((tp->snd_cwnd >> 1) +
728 			    (tp->snd_cwnd >> 2)));
729 }
730 
731 static inline void tcp_sync_left_out(struct tcp_sock *tp)
732 {
733 	if (tp->rx_opt.sack_ok &&
734 	    (tp->sacked_out >= tp->packets_out - tp->lost_out))
735 		tp->sacked_out = tp->packets_out - tp->lost_out;
736 	tp->left_out = tp->sacked_out + tp->lost_out;
737 }
738 
739 extern void tcp_enter_cwr(struct sock *sk);
740 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
741 
742 /* Slow start with delack produces 3 packets of burst, so that
743  * it is safe "de facto".
744  */
745 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
746 {
747 	return 3;
748 }
749 
750 /* RFC2861 Check whether we are limited by application or congestion window
751  * This is the inverse of cwnd check in tcp_tso_should_defer
752  */
753 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
754 {
755 	const struct tcp_sock *tp = tcp_sk(sk);
756 	u32 left;
757 
758 	if (in_flight >= tp->snd_cwnd)
759 		return 1;
760 
761 	if (!sk_can_gso(sk))
762 		return 0;
763 
764 	left = tp->snd_cwnd - in_flight;
765 	if (sysctl_tcp_tso_win_divisor)
766 		return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
767 	else
768 		return left <= tcp_max_burst(tp);
769 }
770 
771 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
772 				       const struct sk_buff *skb)
773 {
774 	if (skb->len < mss)
775 		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
776 }
777 
778 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
779 {
780 	const struct inet_connection_sock *icsk = inet_csk(sk);
781 	if (!tp->packets_out && !icsk->icsk_pending)
782 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
783 					  icsk->icsk_rto, TCP_RTO_MAX);
784 }
785 
786 static inline void tcp_push_pending_frames(struct sock *sk,
787 					   struct tcp_sock *tp)
788 {
789 	__tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
790 }
791 
792 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
793 {
794 	tp->snd_wl1 = seq;
795 }
796 
797 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
798 {
799 	tp->snd_wl1 = seq;
800 }
801 
802 /*
803  * Calculate(/check) TCP checksum
804  */
805 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
806 				   __be32 daddr, __wsum base)
807 {
808 	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
809 }
810 
811 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
812 {
813 	return __skb_checksum_complete(skb);
814 }
815 
816 static inline int tcp_checksum_complete(struct sk_buff *skb)
817 {
818 	return skb->ip_summed != CHECKSUM_UNNECESSARY &&
819 		__tcp_checksum_complete(skb);
820 }
821 
822 /* Prequeue for VJ style copy to user, combined with checksumming. */
823 
824 static inline void tcp_prequeue_init(struct tcp_sock *tp)
825 {
826 	tp->ucopy.task = NULL;
827 	tp->ucopy.len = 0;
828 	tp->ucopy.memory = 0;
829 	skb_queue_head_init(&tp->ucopy.prequeue);
830 #ifdef CONFIG_NET_DMA
831 	tp->ucopy.dma_chan = NULL;
832 	tp->ucopy.wakeup = 0;
833 	tp->ucopy.pinned_list = NULL;
834 	tp->ucopy.dma_cookie = 0;
835 #endif
836 }
837 
838 /* Packet is added to VJ-style prequeue for processing in process
839  * context, if a reader task is waiting. Apparently, this exciting
840  * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
841  * failed somewhere. Latency? Burstiness? Well, at least now we will
842  * see, why it failed. 8)8)				  --ANK
843  *
844  * NOTE: is this not too big to inline?
845  */
846 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
847 {
848 	struct tcp_sock *tp = tcp_sk(sk);
849 
850 	if (!sysctl_tcp_low_latency && tp->ucopy.task) {
851 		__skb_queue_tail(&tp->ucopy.prequeue, skb);
852 		tp->ucopy.memory += skb->truesize;
853 		if (tp->ucopy.memory > sk->sk_rcvbuf) {
854 			struct sk_buff *skb1;
855 
856 			BUG_ON(sock_owned_by_user(sk));
857 
858 			while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
859 				sk->sk_backlog_rcv(sk, skb1);
860 				NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
861 			}
862 
863 			tp->ucopy.memory = 0;
864 		} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
865 			wake_up_interruptible(sk->sk_sleep);
866 			if (!inet_csk_ack_scheduled(sk))
867 				inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
868 						          (3 * TCP_RTO_MIN) / 4,
869 							  TCP_RTO_MAX);
870 		}
871 		return 1;
872 	}
873 	return 0;
874 }
875 
876 
877 #undef STATE_TRACE
878 
879 #ifdef STATE_TRACE
880 static const char *statename[]={
881 	"Unused","Established","Syn Sent","Syn Recv",
882 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
883 	"Close Wait","Last ACK","Listen","Closing"
884 };
885 #endif
886 
887 static inline void tcp_set_state(struct sock *sk, int state)
888 {
889 	int oldstate = sk->sk_state;
890 
891 	switch (state) {
892 	case TCP_ESTABLISHED:
893 		if (oldstate != TCP_ESTABLISHED)
894 			TCP_INC_STATS(TCP_MIB_CURRESTAB);
895 		break;
896 
897 	case TCP_CLOSE:
898 		if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
899 			TCP_INC_STATS(TCP_MIB_ESTABRESETS);
900 
901 		sk->sk_prot->unhash(sk);
902 		if (inet_csk(sk)->icsk_bind_hash &&
903 		    !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
904 			inet_put_port(&tcp_hashinfo, sk);
905 		/* fall through */
906 	default:
907 		if (oldstate==TCP_ESTABLISHED)
908 			TCP_DEC_STATS(TCP_MIB_CURRESTAB);
909 	}
910 
911 	/* Change state AFTER socket is unhashed to avoid closed
912 	 * socket sitting in hash tables.
913 	 */
914 	sk->sk_state = state;
915 
916 #ifdef STATE_TRACE
917 	SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
918 #endif
919 }
920 
921 static inline void tcp_done(struct sock *sk)
922 {
923 	if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
924 		TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
925 
926 	tcp_set_state(sk, TCP_CLOSE);
927 	tcp_clear_xmit_timers(sk);
928 
929 	sk->sk_shutdown = SHUTDOWN_MASK;
930 
931 	if (!sock_flag(sk, SOCK_DEAD))
932 		sk->sk_state_change(sk);
933 	else
934 		inet_csk_destroy_sock(sk);
935 }
936 
937 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
938 {
939 	rx_opt->dsack = 0;
940 	rx_opt->eff_sacks = 0;
941 	rx_opt->num_sacks = 0;
942 }
943 
944 /* Determine a window scaling and initial window to offer. */
945 extern void tcp_select_initial_window(int __space, __u32 mss,
946 				      __u32 *rcv_wnd, __u32 *window_clamp,
947 				      int wscale_ok, __u8 *rcv_wscale);
948 
949 static inline int tcp_win_from_space(int space)
950 {
951 	return sysctl_tcp_adv_win_scale<=0 ?
952 		(space>>(-sysctl_tcp_adv_win_scale)) :
953 		space - (space>>sysctl_tcp_adv_win_scale);
954 }
955 
956 /* Note: caller must be prepared to deal with negative returns */
957 static inline int tcp_space(const struct sock *sk)
958 {
959 	return tcp_win_from_space(sk->sk_rcvbuf -
960 				  atomic_read(&sk->sk_rmem_alloc));
961 }
962 
963 static inline int tcp_full_space(const struct sock *sk)
964 {
965 	return tcp_win_from_space(sk->sk_rcvbuf);
966 }
967 
968 static inline void tcp_openreq_init(struct request_sock *req,
969 				    struct tcp_options_received *rx_opt,
970 				    struct sk_buff *skb)
971 {
972 	struct inet_request_sock *ireq = inet_rsk(req);
973 
974 	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
975 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
976 	req->mss = rx_opt->mss_clamp;
977 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
978 	ireq->tstamp_ok = rx_opt->tstamp_ok;
979 	ireq->sack_ok = rx_opt->sack_ok;
980 	ireq->snd_wscale = rx_opt->snd_wscale;
981 	ireq->wscale_ok = rx_opt->wscale_ok;
982 	ireq->acked = 0;
983 	ireq->ecn_ok = 0;
984 	ireq->rmt_port = skb->h.th->source;
985 }
986 
987 extern void tcp_enter_memory_pressure(void);
988 
989 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
990 {
991 	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
992 }
993 
994 static inline int keepalive_time_when(const struct tcp_sock *tp)
995 {
996 	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
997 }
998 
999 static inline int tcp_fin_time(const struct sock *sk)
1000 {
1001 	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1002 	const int rto = inet_csk(sk)->icsk_rto;
1003 
1004 	if (fin_timeout < (rto << 2) - (rto >> 1))
1005 		fin_timeout = (rto << 2) - (rto >> 1);
1006 
1007 	return fin_timeout;
1008 }
1009 
1010 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1011 {
1012 	if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1013 		return 0;
1014 	if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1015 		return 0;
1016 
1017 	/* RST segments are not recommended to carry timestamp,
1018 	   and, if they do, it is recommended to ignore PAWS because
1019 	   "their cleanup function should take precedence over timestamps."
1020 	   Certainly, it is mistake. It is necessary to understand the reasons
1021 	   of this constraint to relax it: if peer reboots, clock may go
1022 	   out-of-sync and half-open connections will not be reset.
1023 	   Actually, the problem would be not existing if all
1024 	   the implementations followed draft about maintaining clock
1025 	   via reboots. Linux-2.2 DOES NOT!
1026 
1027 	   However, we can relax time bounds for RST segments to MSL.
1028 	 */
1029 	if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1030 		return 0;
1031 	return 1;
1032 }
1033 
1034 #define TCP_CHECK_TIMER(sk) do { } while (0)
1035 
1036 static inline int tcp_use_frto(const struct sock *sk)
1037 {
1038 	const struct tcp_sock *tp = tcp_sk(sk);
1039 
1040 	/* F-RTO must be activated in sysctl and there must be some
1041 	 * unsent new data, and the advertised window should allow
1042 	 * sending it.
1043 	 */
1044 	return (sysctl_tcp_frto && sk->sk_send_head &&
1045 		!after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1046 		       tp->snd_una + tp->snd_wnd));
1047 }
1048 
1049 static inline void tcp_mib_init(void)
1050 {
1051 	/* See RFC 2012 */
1052 	TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1053 	TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1054 	TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1055 	TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1056 }
1057 
1058 /*from STCP */
1059 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1060 	tp->lost_skb_hint = NULL;
1061 	tp->scoreboard_skb_hint = NULL;
1062 	tp->retransmit_skb_hint = NULL;
1063 	tp->forward_skb_hint = NULL;
1064 	tp->fastpath_skb_hint = NULL;
1065 }
1066 
1067 /* MD5 Signature */
1068 struct crypto_hash;
1069 
1070 /* - key database */
1071 struct tcp_md5sig_key {
1072 	u8			*key;
1073 	u8			keylen;
1074 };
1075 
1076 struct tcp4_md5sig_key {
1077 	u8			*key;
1078 	u16			keylen;
1079 	__be32			addr;
1080 };
1081 
1082 struct tcp6_md5sig_key {
1083 	u8			*key;
1084 	u16			keylen;
1085 #if 0
1086 	u32			scope_id;	/* XXX */
1087 #endif
1088 	struct in6_addr		addr;
1089 };
1090 
1091 /* - sock block */
1092 struct tcp_md5sig_info {
1093 	struct tcp4_md5sig_key	*keys4;
1094 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1095 	struct tcp6_md5sig_key	*keys6;
1096 	u32			entries6;
1097 	u32			alloced6;
1098 #endif
1099 	u32			entries4;
1100 	u32			alloced4;
1101 };
1102 
1103 /* - pseudo header */
1104 struct tcp4_pseudohdr {
1105 	__be32		saddr;
1106 	__be32		daddr;
1107 	__u8		pad;
1108 	__u8		protocol;
1109 	__be16		len;
1110 };
1111 
1112 struct tcp6_pseudohdr {
1113 	struct in6_addr	saddr;
1114 	struct in6_addr daddr;
1115 	__be32		len;
1116 	__be32		protocol;	/* including padding */
1117 };
1118 
1119 union tcp_md5sum_block {
1120 	struct tcp4_pseudohdr ip4;
1121 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1122 	struct tcp6_pseudohdr ip6;
1123 #endif
1124 };
1125 
1126 /* - pool: digest algorithm, hash description and scratch buffer */
1127 struct tcp_md5sig_pool {
1128 	struct hash_desc	md5_desc;
1129 	union tcp_md5sum_block	md5_blk;
1130 };
1131 
1132 #define TCP_MD5SIG_MAXKEYS	(~(u32)0)	/* really?! */
1133 
1134 /* - functions */
1135 extern int			tcp_v4_calc_md5_hash(char *md5_hash,
1136 						     struct tcp_md5sig_key *key,
1137 						     struct sock *sk,
1138 						     struct dst_entry *dst,
1139 						     struct request_sock *req,
1140 						     struct tcphdr *th,
1141 						     int protocol, int tcplen);
1142 extern struct tcp_md5sig_key	*tcp_v4_md5_lookup(struct sock *sk,
1143 						   struct sock *addr_sk);
1144 
1145 extern int			tcp_v4_md5_do_add(struct sock *sk,
1146 						  __be32 addr,
1147 						  u8 *newkey,
1148 						  u8 newkeylen);
1149 
1150 extern int			tcp_v4_md5_do_del(struct sock *sk,
1151 						  __be32 addr);
1152 
1153 extern struct tcp_md5sig_pool	**tcp_alloc_md5sig_pool(void);
1154 extern void			tcp_free_md5sig_pool(void);
1155 
1156 extern struct tcp_md5sig_pool	*__tcp_get_md5sig_pool(int cpu);
1157 extern void			__tcp_put_md5sig_pool(void);
1158 
1159 static inline
1160 struct tcp_md5sig_pool		*tcp_get_md5sig_pool(void)
1161 {
1162 	int cpu = get_cpu();
1163 	struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1164 	if (!ret)
1165 		put_cpu();
1166 	return ret;
1167 }
1168 
1169 static inline void		tcp_put_md5sig_pool(void)
1170 {
1171 	__tcp_put_md5sig_pool();
1172 	put_cpu();
1173 }
1174 
1175 /* /proc */
1176 enum tcp_seq_states {
1177 	TCP_SEQ_STATE_LISTENING,
1178 	TCP_SEQ_STATE_OPENREQ,
1179 	TCP_SEQ_STATE_ESTABLISHED,
1180 	TCP_SEQ_STATE_TIME_WAIT,
1181 };
1182 
1183 struct tcp_seq_afinfo {
1184 	struct module		*owner;
1185 	char			*name;
1186 	sa_family_t		family;
1187 	int			(*seq_show) (struct seq_file *m, void *v);
1188 	struct file_operations	*seq_fops;
1189 };
1190 
1191 struct tcp_iter_state {
1192 	sa_family_t		family;
1193 	enum tcp_seq_states	state;
1194 	struct sock		*syn_wait_sk;
1195 	int			bucket, sbucket, num, uid;
1196 	struct seq_operations	seq_ops;
1197 };
1198 
1199 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1200 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1201 
1202 extern struct request_sock_ops tcp_request_sock_ops;
1203 
1204 extern int tcp_v4_destroy_sock(struct sock *sk);
1205 
1206 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1207 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1208 
1209 #ifdef CONFIG_PROC_FS
1210 extern int  tcp4_proc_init(void);
1211 extern void tcp4_proc_exit(void);
1212 #endif
1213 
1214 /* TCP af-specific functions */
1215 struct tcp_sock_af_ops {
1216 #ifdef CONFIG_TCP_MD5SIG
1217 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1218 						struct sock *addr_sk);
1219 	int			(*calc_md5_hash) (char *location,
1220 						  struct tcp_md5sig_key *md5,
1221 						  struct sock *sk,
1222 						  struct dst_entry *dst,
1223 						  struct request_sock *req,
1224 						  struct tcphdr *th,
1225 						  int protocol, int len);
1226 	int			(*md5_add) (struct sock *sk,
1227 					    struct sock *addr_sk,
1228 					    u8 *newkey,
1229 					    u8 len);
1230 	int			(*md5_parse) (struct sock *sk,
1231 					      char __user *optval,
1232 					      int optlen);
1233 #endif
1234 };
1235 
1236 struct tcp_request_sock_ops {
1237 #ifdef CONFIG_TCP_MD5SIG
1238 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1239 						struct request_sock *req);
1240 #endif
1241 };
1242 
1243 extern void tcp_v4_init(struct net_proto_family *ops);
1244 extern void tcp_init(void);
1245 
1246 #endif	/* _TCP_H */
1247