xref: /linux/include/net/tcp.h (revision aeb3f46252e26acdc60a1a8e31fb1ca6319d9a07)
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_frto_response;
224 extern int sysctl_tcp_low_latency;
225 extern int sysctl_tcp_dma_copybreak;
226 extern int sysctl_tcp_nometrics_save;
227 extern int sysctl_tcp_moderate_rcvbuf;
228 extern int sysctl_tcp_tso_win_divisor;
229 extern int sysctl_tcp_abc;
230 extern int sysctl_tcp_mtu_probing;
231 extern int sysctl_tcp_base_mss;
232 extern int sysctl_tcp_workaround_signed_windows;
233 extern int sysctl_tcp_slow_start_after_idle;
234 extern int sysctl_tcp_max_ssthresh;
235 
236 extern atomic_t tcp_memory_allocated;
237 extern atomic_t tcp_sockets_allocated;
238 extern int tcp_memory_pressure;
239 
240 /*
241  * The next routines deal with comparing 32 bit unsigned ints
242  * and worry about wraparound (automatic with unsigned arithmetic).
243  */
244 
245 static inline int before(__u32 seq1, __u32 seq2)
246 {
247         return (__s32)(seq1-seq2) < 0;
248 }
249 #define after(seq2, seq1) 	before(seq1, seq2)
250 
251 /* is s2<=s1<=s3 ? */
252 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
253 {
254 	return seq3 - seq2 >= seq1 - seq2;
255 }
256 
257 static inline int tcp_too_many_orphans(struct sock *sk, int num)
258 {
259 	return (num > sysctl_tcp_max_orphans) ||
260 		(sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
261 		 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
262 }
263 
264 extern struct proto tcp_prot;
265 
266 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
267 #define TCP_INC_STATS(field)		SNMP_INC_STATS(tcp_statistics, field)
268 #define TCP_INC_STATS_BH(field)		SNMP_INC_STATS_BH(tcp_statistics, field)
269 #define TCP_INC_STATS_USER(field) 	SNMP_INC_STATS_USER(tcp_statistics, field)
270 #define TCP_DEC_STATS(field)		SNMP_DEC_STATS(tcp_statistics, field)
271 #define TCP_ADD_STATS_BH(field, val)	SNMP_ADD_STATS_BH(tcp_statistics, field, val)
272 #define TCP_ADD_STATS_USER(field, val)	SNMP_ADD_STATS_USER(tcp_statistics, field, val)
273 
274 extern void			tcp_v4_err(struct sk_buff *skb, u32);
275 
276 extern void			tcp_shutdown (struct sock *sk, int how);
277 
278 extern int			tcp_v4_rcv(struct sk_buff *skb);
279 
280 extern int			tcp_v4_remember_stamp(struct sock *sk);
281 
282 extern int		    	tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
283 
284 extern int			tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
285 					    struct msghdr *msg, size_t size);
286 extern ssize_t			tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
287 
288 extern int			tcp_ioctl(struct sock *sk,
289 					  int cmd,
290 					  unsigned long arg);
291 
292 extern int			tcp_rcv_state_process(struct sock *sk,
293 						      struct sk_buff *skb,
294 						      struct tcphdr *th,
295 						      unsigned len);
296 
297 extern int			tcp_rcv_established(struct sock *sk,
298 						    struct sk_buff *skb,
299 						    struct tcphdr *th,
300 						    unsigned len);
301 
302 extern void			tcp_rcv_space_adjust(struct sock *sk);
303 
304 extern void			tcp_cleanup_rbuf(struct sock *sk, int copied);
305 
306 extern int			tcp_twsk_unique(struct sock *sk,
307 						struct sock *sktw, void *twp);
308 
309 extern void			tcp_twsk_destructor(struct sock *sk);
310 
311 static inline void tcp_dec_quickack_mode(struct sock *sk,
312 					 const unsigned int pkts)
313 {
314 	struct inet_connection_sock *icsk = inet_csk(sk);
315 
316 	if (icsk->icsk_ack.quick) {
317 		if (pkts >= icsk->icsk_ack.quick) {
318 			icsk->icsk_ack.quick = 0;
319 			/* Leaving quickack mode we deflate ATO. */
320 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
321 		} else
322 			icsk->icsk_ack.quick -= pkts;
323 	}
324 }
325 
326 extern void tcp_enter_quickack_mode(struct sock *sk);
327 
328 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
329 {
330  	rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
331 }
332 
333 enum tcp_tw_status
334 {
335 	TCP_TW_SUCCESS = 0,
336 	TCP_TW_RST = 1,
337 	TCP_TW_ACK = 2,
338 	TCP_TW_SYN = 3
339 };
340 
341 
342 extern enum tcp_tw_status	tcp_timewait_state_process(struct inet_timewait_sock *tw,
343 							   struct sk_buff *skb,
344 							   const struct tcphdr *th);
345 
346 extern struct sock *		tcp_check_req(struct sock *sk,struct sk_buff *skb,
347 					      struct request_sock *req,
348 					      struct request_sock **prev);
349 extern int			tcp_child_process(struct sock *parent,
350 						  struct sock *child,
351 						  struct sk_buff *skb);
352 extern int			tcp_use_frto(struct sock *sk);
353 extern void			tcp_enter_frto(struct sock *sk);
354 extern void			tcp_enter_loss(struct sock *sk, int how);
355 extern void			tcp_clear_retrans(struct tcp_sock *tp);
356 extern void			tcp_update_metrics(struct sock *sk);
357 
358 extern void			tcp_close(struct sock *sk,
359 					  long timeout);
360 extern unsigned int		tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
361 
362 extern int			tcp_getsockopt(struct sock *sk, int level,
363 					       int optname,
364 					       char __user *optval,
365 					       int __user *optlen);
366 extern int			tcp_setsockopt(struct sock *sk, int level,
367 					       int optname, char __user *optval,
368 					       int optlen);
369 extern int			compat_tcp_getsockopt(struct sock *sk,
370 					int level, int optname,
371 					char __user *optval, int __user *optlen);
372 extern int			compat_tcp_setsockopt(struct sock *sk,
373 					int level, int optname,
374 					char __user *optval, int optlen);
375 extern void			tcp_set_keepalive(struct sock *sk, int val);
376 extern int			tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
377 					    struct msghdr *msg,
378 					    size_t len, int nonblock,
379 					    int flags, int *addr_len);
380 
381 extern void			tcp_parse_options(struct sk_buff *skb,
382 						  struct tcp_options_received *opt_rx,
383 						  int estab);
384 
385 /*
386  *	TCP v4 functions exported for the inet6 API
387  */
388 
389 extern void		       	tcp_v4_send_check(struct sock *sk, int len,
390 						  struct sk_buff *skb);
391 
392 extern int			tcp_v4_conn_request(struct sock *sk,
393 						    struct sk_buff *skb);
394 
395 extern struct sock *		tcp_create_openreq_child(struct sock *sk,
396 							 struct request_sock *req,
397 							 struct sk_buff *skb);
398 
399 extern struct sock *		tcp_v4_syn_recv_sock(struct sock *sk,
400 						     struct sk_buff *skb,
401 						     struct request_sock *req,
402 							struct dst_entry *dst);
403 
404 extern int			tcp_v4_do_rcv(struct sock *sk,
405 					      struct sk_buff *skb);
406 
407 extern int			tcp_v4_connect(struct sock *sk,
408 					       struct sockaddr *uaddr,
409 					       int addr_len);
410 
411 extern int			tcp_connect(struct sock *sk);
412 
413 extern struct sk_buff *		tcp_make_synack(struct sock *sk,
414 						struct dst_entry *dst,
415 						struct request_sock *req);
416 
417 extern int			tcp_disconnect(struct sock *sk, int flags);
418 
419 extern void			tcp_unhash(struct sock *sk);
420 
421 /* From syncookies.c */
422 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
423 				    struct ip_options *opt);
424 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
425 				     __u16 *mss);
426 
427 /* tcp_output.c */
428 
429 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
430 				      int nonagle);
431 extern int tcp_may_send_now(struct sock *sk);
432 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
433 extern void tcp_xmit_retransmit_queue(struct sock *);
434 extern void tcp_simple_retransmit(struct sock *);
435 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
436 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
437 
438 extern void tcp_send_probe0(struct sock *);
439 extern void tcp_send_partial(struct sock *);
440 extern int  tcp_write_wakeup(struct sock *);
441 extern void tcp_send_fin(struct sock *sk);
442 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
443 extern int  tcp_send_synack(struct sock *);
444 extern void tcp_push_one(struct sock *, unsigned int mss_now);
445 extern void tcp_send_ack(struct sock *sk);
446 extern void tcp_send_delayed_ack(struct sock *sk);
447 
448 /* tcp_input.c */
449 extern void tcp_cwnd_application_limited(struct sock *sk);
450 
451 /* tcp_timer.c */
452 extern void tcp_init_xmit_timers(struct sock *);
453 static inline void tcp_clear_xmit_timers(struct sock *sk)
454 {
455 	inet_csk_clear_xmit_timers(sk);
456 }
457 
458 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
459 extern unsigned int tcp_current_mss(struct sock *sk, int large);
460 
461 /* tcp.c */
462 extern void tcp_get_info(struct sock *, struct tcp_info *);
463 
464 /* Read 'sendfile()'-style from a TCP socket */
465 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
466 				unsigned int, size_t);
467 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
468 			 sk_read_actor_t recv_actor);
469 
470 extern void tcp_initialize_rcv_mss(struct sock *sk);
471 
472 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
473 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
474 extern void tcp_mtup_init(struct sock *sk);
475 
476 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
477 {
478 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
479 			       ntohl(TCP_FLAG_ACK) |
480 			       snd_wnd);
481 }
482 
483 static inline void tcp_fast_path_on(struct tcp_sock *tp)
484 {
485 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
486 }
487 
488 static inline void tcp_fast_path_check(struct sock *sk)
489 {
490 	struct tcp_sock *tp = tcp_sk(sk);
491 
492 	if (skb_queue_empty(&tp->out_of_order_queue) &&
493 	    tp->rcv_wnd &&
494 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
495 	    !tp->urg_data)
496 		tcp_fast_path_on(tp);
497 }
498 
499 /* Compute the actual receive window we are currently advertising.
500  * Rcv_nxt can be after the window if our peer push more data
501  * than the offered window.
502  */
503 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
504 {
505 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
506 
507 	if (win < 0)
508 		win = 0;
509 	return (u32) win;
510 }
511 
512 /* Choose a new window, without checks for shrinking, and without
513  * scaling applied to the result.  The caller does these things
514  * if necessary.  This is a "raw" window selection.
515  */
516 extern u32	__tcp_select_window(struct sock *sk);
517 
518 /* TCP timestamps are only 32-bits, this causes a slight
519  * complication on 64-bit systems since we store a snapshot
520  * of jiffies in the buffer control blocks below.  We decided
521  * to use only the low 32-bits of jiffies and hide the ugly
522  * casts with the following macro.
523  */
524 #define tcp_time_stamp		((__u32)(jiffies))
525 
526 /* This is what the send packet queuing engine uses to pass
527  * TCP per-packet control information to the transmission
528  * code.  We also store the host-order sequence numbers in
529  * here too.  This is 36 bytes on 32-bit architectures,
530  * 40 bytes on 64-bit machines, if this grows please adjust
531  * skbuff.h:skbuff->cb[xxx] size appropriately.
532  */
533 struct tcp_skb_cb {
534 	union {
535 		struct inet_skb_parm	h4;
536 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
537 		struct inet6_skb_parm	h6;
538 #endif
539 	} header;	/* For incoming frames		*/
540 	__u32		seq;		/* Starting sequence number	*/
541 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
542 	__u32		when;		/* used to compute rtt's	*/
543 	__u8		flags;		/* TCP header flags.		*/
544 
545 	/* NOTE: These must match up to the flags byte in a
546 	 *       real TCP header.
547 	 */
548 #define TCPCB_FLAG_FIN		0x01
549 #define TCPCB_FLAG_SYN		0x02
550 #define TCPCB_FLAG_RST		0x04
551 #define TCPCB_FLAG_PSH		0x08
552 #define TCPCB_FLAG_ACK		0x10
553 #define TCPCB_FLAG_URG		0x20
554 #define TCPCB_FLAG_ECE		0x40
555 #define TCPCB_FLAG_CWR		0x80
556 
557 	__u8		sacked;		/* State flags for SACK/FACK.	*/
558 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
559 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
560 #define TCPCB_LOST		0x04	/* SKB is lost			*/
561 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
562 
563 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
564 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
565 
566 #define TCPCB_URG		0x20	/* Urgent pointer advanced here	*/
567 
568 #define TCPCB_AT_TAIL		(TCPCB_URG)
569 
570 	__u16		urg_ptr;	/* Valid w/URG flags is set.	*/
571 	__u32		ack_seq;	/* Sequence number ACK'd	*/
572 };
573 
574 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
575 
576 #include <net/tcp_ecn.h>
577 
578 /* Due to TSO, an SKB can be composed of multiple actual
579  * packets.  To keep these tracked properly, we use this.
580  */
581 static inline int tcp_skb_pcount(const struct sk_buff *skb)
582 {
583 	return skb_shinfo(skb)->gso_segs;
584 }
585 
586 /* This is valid iff tcp_skb_pcount() > 1. */
587 static inline int tcp_skb_mss(const struct sk_buff *skb)
588 {
589 	return skb_shinfo(skb)->gso_size;
590 }
591 
592 static inline void tcp_dec_pcount_approx(__u32 *count,
593 					 const struct sk_buff *skb)
594 {
595 	if (*count) {
596 		*count -= tcp_skb_pcount(skb);
597 		if ((int)*count < 0)
598 			*count = 0;
599 	}
600 }
601 
602 static inline void tcp_packets_out_inc(struct sock *sk,
603 				       const struct sk_buff *skb)
604 {
605 	struct tcp_sock *tp = tcp_sk(sk);
606 	int orig = tp->packets_out;
607 
608 	tp->packets_out += tcp_skb_pcount(skb);
609 	if (!orig)
610 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
611 					  inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
612 }
613 
614 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
615 				       const struct sk_buff *skb)
616 {
617 	tp->packets_out -= tcp_skb_pcount(skb);
618 }
619 
620 /* Events passed to congestion control interface */
621 enum tcp_ca_event {
622 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
623 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
624 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
625 	CA_EVENT_FRTO,		/* fast recovery timeout */
626 	CA_EVENT_LOSS,		/* loss timeout */
627 	CA_EVENT_FAST_ACK,	/* in sequence ack */
628 	CA_EVENT_SLOW_ACK,	/* other ack */
629 };
630 
631 /*
632  * Interface for adding new TCP congestion control handlers
633  */
634 #define TCP_CA_NAME_MAX	16
635 #define TCP_CA_MAX	128
636 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
637 
638 #define TCP_CONG_NON_RESTRICTED 0x1
639 #define TCP_CONG_RTT_STAMP	0x2
640 
641 struct tcp_congestion_ops {
642 	struct list_head	list;
643 	unsigned long flags;
644 
645 	/* initialize private data (optional) */
646 	void (*init)(struct sock *sk);
647 	/* cleanup private data  (optional) */
648 	void (*release)(struct sock *sk);
649 
650 	/* return slow start threshold (required) */
651 	u32 (*ssthresh)(struct sock *sk);
652 	/* lower bound for congestion window (optional) */
653 	u32 (*min_cwnd)(const struct sock *sk);
654 	/* do new cwnd calculation (required) */
655 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight, int good_ack);
656 	/* call before changing ca_state (optional) */
657 	void (*set_state)(struct sock *sk, u8 new_state);
658 	/* call when cwnd event occurs (optional) */
659 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
660 	/* new value of cwnd after loss (optional) */
661 	u32  (*undo_cwnd)(struct sock *sk);
662 	/* hook for packet ack accounting (optional) */
663 	void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last);
664 	/* get info for inet_diag (optional) */
665 	void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
666 
667 	char 		name[TCP_CA_NAME_MAX];
668 	struct module 	*owner;
669 };
670 
671 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
672 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
673 
674 extern void tcp_init_congestion_control(struct sock *sk);
675 extern void tcp_cleanup_congestion_control(struct sock *sk);
676 extern int tcp_set_default_congestion_control(const char *name);
677 extern void tcp_get_default_congestion_control(char *name);
678 extern void tcp_get_available_congestion_control(char *buf, size_t len);
679 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
680 extern int tcp_set_allowed_congestion_control(char *allowed);
681 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
682 extern void tcp_slow_start(struct tcp_sock *tp);
683 
684 extern struct tcp_congestion_ops tcp_init_congestion_ops;
685 extern u32 tcp_reno_ssthresh(struct sock *sk);
686 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight, int flag);
687 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
688 extern struct tcp_congestion_ops tcp_reno;
689 
690 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
691 {
692 	struct inet_connection_sock *icsk = inet_csk(sk);
693 
694 	if (icsk->icsk_ca_ops->set_state)
695 		icsk->icsk_ca_ops->set_state(sk, ca_state);
696 	icsk->icsk_ca_state = ca_state;
697 }
698 
699 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
700 {
701 	const struct inet_connection_sock *icsk = inet_csk(sk);
702 
703 	if (icsk->icsk_ca_ops->cwnd_event)
704 		icsk->icsk_ca_ops->cwnd_event(sk, event);
705 }
706 
707 /* This determines how many packets are "in the network" to the best
708  * of our knowledge.  In many cases it is conservative, but where
709  * detailed information is available from the receiver (via SACK
710  * blocks etc.) we can make more aggressive calculations.
711  *
712  * Use this for decisions involving congestion control, use just
713  * tp->packets_out to determine if the send queue is empty or not.
714  *
715  * Read this equation as:
716  *
717  *	"Packets sent once on transmission queue" MINUS
718  *	"Packets left network, but not honestly ACKed yet" PLUS
719  *	"Packets fast retransmitted"
720  */
721 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
722 {
723 	return (tp->packets_out - tp->left_out + tp->retrans_out);
724 }
725 
726 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
727  * The exception is rate halving phase, when cwnd is decreasing towards
728  * ssthresh.
729  */
730 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
731 {
732 	const struct tcp_sock *tp = tcp_sk(sk);
733 	if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
734 		return tp->snd_ssthresh;
735 	else
736 		return max(tp->snd_ssthresh,
737 			   ((tp->snd_cwnd >> 1) +
738 			    (tp->snd_cwnd >> 2)));
739 }
740 
741 static inline void tcp_sync_left_out(struct tcp_sock *tp)
742 {
743 	BUG_ON(tp->rx_opt.sack_ok &&
744 	       (tp->sacked_out + tp->lost_out > tp->packets_out));
745 	tp->left_out = tp->sacked_out + tp->lost_out;
746 }
747 
748 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
749 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
750 
751 /* Slow start with delack produces 3 packets of burst, so that
752  * it is safe "de facto".
753  */
754 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
755 {
756 	return 3;
757 }
758 
759 /* RFC2861 Check whether we are limited by application or congestion window
760  * This is the inverse of cwnd check in tcp_tso_should_defer
761  */
762 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
763 {
764 	const struct tcp_sock *tp = tcp_sk(sk);
765 	u32 left;
766 
767 	if (in_flight >= tp->snd_cwnd)
768 		return 1;
769 
770 	if (!sk_can_gso(sk))
771 		return 0;
772 
773 	left = tp->snd_cwnd - in_flight;
774 	if (sysctl_tcp_tso_win_divisor)
775 		return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
776 	else
777 		return left <= tcp_max_burst(tp);
778 }
779 
780 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
781 				       const struct sk_buff *skb)
782 {
783 	if (skb->len < mss)
784 		tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
785 }
786 
787 static inline void tcp_check_probe_timer(struct sock *sk)
788 {
789 	struct tcp_sock *tp = tcp_sk(sk);
790 	const struct inet_connection_sock *icsk = inet_csk(sk);
791 
792 	if (!tp->packets_out && !icsk->icsk_pending)
793 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
794 					  icsk->icsk_rto, TCP_RTO_MAX);
795 }
796 
797 static inline void tcp_push_pending_frames(struct sock *sk)
798 {
799 	struct tcp_sock *tp = tcp_sk(sk);
800 
801 	__tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
802 }
803 
804 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
805 {
806 	tp->snd_wl1 = seq;
807 }
808 
809 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
810 {
811 	tp->snd_wl1 = seq;
812 }
813 
814 /*
815  * Calculate(/check) TCP checksum
816  */
817 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
818 				   __be32 daddr, __wsum base)
819 {
820 	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
821 }
822 
823 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
824 {
825 	return __skb_checksum_complete(skb);
826 }
827 
828 static inline int tcp_checksum_complete(struct sk_buff *skb)
829 {
830 	return !skb_csum_unnecessary(skb) &&
831 		__tcp_checksum_complete(skb);
832 }
833 
834 /* Prequeue for VJ style copy to user, combined with checksumming. */
835 
836 static inline void tcp_prequeue_init(struct tcp_sock *tp)
837 {
838 	tp->ucopy.task = NULL;
839 	tp->ucopy.len = 0;
840 	tp->ucopy.memory = 0;
841 	skb_queue_head_init(&tp->ucopy.prequeue);
842 #ifdef CONFIG_NET_DMA
843 	tp->ucopy.dma_chan = NULL;
844 	tp->ucopy.wakeup = 0;
845 	tp->ucopy.pinned_list = NULL;
846 	tp->ucopy.dma_cookie = 0;
847 #endif
848 }
849 
850 /* Packet is added to VJ-style prequeue for processing in process
851  * context, if a reader task is waiting. Apparently, this exciting
852  * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
853  * failed somewhere. Latency? Burstiness? Well, at least now we will
854  * see, why it failed. 8)8)				  --ANK
855  *
856  * NOTE: is this not too big to inline?
857  */
858 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
859 {
860 	struct tcp_sock *tp = tcp_sk(sk);
861 
862 	if (!sysctl_tcp_low_latency && tp->ucopy.task) {
863 		__skb_queue_tail(&tp->ucopy.prequeue, skb);
864 		tp->ucopy.memory += skb->truesize;
865 		if (tp->ucopy.memory > sk->sk_rcvbuf) {
866 			struct sk_buff *skb1;
867 
868 			BUG_ON(sock_owned_by_user(sk));
869 
870 			while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
871 				sk->sk_backlog_rcv(sk, skb1);
872 				NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
873 			}
874 
875 			tp->ucopy.memory = 0;
876 		} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
877 			wake_up_interruptible(sk->sk_sleep);
878 			if (!inet_csk_ack_scheduled(sk))
879 				inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
880 						          (3 * TCP_RTO_MIN) / 4,
881 							  TCP_RTO_MAX);
882 		}
883 		return 1;
884 	}
885 	return 0;
886 }
887 
888 
889 #undef STATE_TRACE
890 
891 #ifdef STATE_TRACE
892 static const char *statename[]={
893 	"Unused","Established","Syn Sent","Syn Recv",
894 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
895 	"Close Wait","Last ACK","Listen","Closing"
896 };
897 #endif
898 
899 static inline void tcp_set_state(struct sock *sk, int state)
900 {
901 	int oldstate = sk->sk_state;
902 
903 	switch (state) {
904 	case TCP_ESTABLISHED:
905 		if (oldstate != TCP_ESTABLISHED)
906 			TCP_INC_STATS(TCP_MIB_CURRESTAB);
907 		break;
908 
909 	case TCP_CLOSE:
910 		if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
911 			TCP_INC_STATS(TCP_MIB_ESTABRESETS);
912 
913 		sk->sk_prot->unhash(sk);
914 		if (inet_csk(sk)->icsk_bind_hash &&
915 		    !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
916 			inet_put_port(&tcp_hashinfo, sk);
917 		/* fall through */
918 	default:
919 		if (oldstate==TCP_ESTABLISHED)
920 			TCP_DEC_STATS(TCP_MIB_CURRESTAB);
921 	}
922 
923 	/* Change state AFTER socket is unhashed to avoid closed
924 	 * socket sitting in hash tables.
925 	 */
926 	sk->sk_state = state;
927 
928 #ifdef STATE_TRACE
929 	SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
930 #endif
931 }
932 
933 extern void tcp_done(struct sock *sk);
934 
935 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
936 {
937 	rx_opt->dsack = 0;
938 	rx_opt->eff_sacks = 0;
939 	rx_opt->num_sacks = 0;
940 }
941 
942 /* Determine a window scaling and initial window to offer. */
943 extern void tcp_select_initial_window(int __space, __u32 mss,
944 				      __u32 *rcv_wnd, __u32 *window_clamp,
945 				      int wscale_ok, __u8 *rcv_wscale);
946 
947 static inline int tcp_win_from_space(int space)
948 {
949 	return sysctl_tcp_adv_win_scale<=0 ?
950 		(space>>(-sysctl_tcp_adv_win_scale)) :
951 		space - (space>>sysctl_tcp_adv_win_scale);
952 }
953 
954 /* Note: caller must be prepared to deal with negative returns */
955 static inline int tcp_space(const struct sock *sk)
956 {
957 	return tcp_win_from_space(sk->sk_rcvbuf -
958 				  atomic_read(&sk->sk_rmem_alloc));
959 }
960 
961 static inline int tcp_full_space(const struct sock *sk)
962 {
963 	return tcp_win_from_space(sk->sk_rcvbuf);
964 }
965 
966 static inline void tcp_openreq_init(struct request_sock *req,
967 				    struct tcp_options_received *rx_opt,
968 				    struct sk_buff *skb)
969 {
970 	struct inet_request_sock *ireq = inet_rsk(req);
971 
972 	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
973 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
974 	req->mss = rx_opt->mss_clamp;
975 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
976 	ireq->tstamp_ok = rx_opt->tstamp_ok;
977 	ireq->sack_ok = rx_opt->sack_ok;
978 	ireq->snd_wscale = rx_opt->snd_wscale;
979 	ireq->wscale_ok = rx_opt->wscale_ok;
980 	ireq->acked = 0;
981 	ireq->ecn_ok = 0;
982 	ireq->rmt_port = tcp_hdr(skb)->source;
983 }
984 
985 extern void tcp_enter_memory_pressure(void);
986 
987 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
988 {
989 	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
990 }
991 
992 static inline int keepalive_time_when(const struct tcp_sock *tp)
993 {
994 	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
995 }
996 
997 static inline int tcp_fin_time(const struct sock *sk)
998 {
999 	int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1000 	const int rto = inet_csk(sk)->icsk_rto;
1001 
1002 	if (fin_timeout < (rto << 2) - (rto >> 1))
1003 		fin_timeout = (rto << 2) - (rto >> 1);
1004 
1005 	return fin_timeout;
1006 }
1007 
1008 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1009 {
1010 	if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1011 		return 0;
1012 	if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1013 		return 0;
1014 
1015 	/* RST segments are not recommended to carry timestamp,
1016 	   and, if they do, it is recommended to ignore PAWS because
1017 	   "their cleanup function should take precedence over timestamps."
1018 	   Certainly, it is mistake. It is necessary to understand the reasons
1019 	   of this constraint to relax it: if peer reboots, clock may go
1020 	   out-of-sync and half-open connections will not be reset.
1021 	   Actually, the problem would be not existing if all
1022 	   the implementations followed draft about maintaining clock
1023 	   via reboots. Linux-2.2 DOES NOT!
1024 
1025 	   However, we can relax time bounds for RST segments to MSL.
1026 	 */
1027 	if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1028 		return 0;
1029 	return 1;
1030 }
1031 
1032 #define TCP_CHECK_TIMER(sk) do { } while (0)
1033 
1034 static inline void tcp_mib_init(void)
1035 {
1036 	/* See RFC 2012 */
1037 	TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1038 	TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1039 	TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1040 	TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1041 }
1042 
1043 /*from STCP */
1044 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1045 	tp->lost_skb_hint = NULL;
1046 	tp->scoreboard_skb_hint = NULL;
1047 	tp->retransmit_skb_hint = NULL;
1048 	tp->forward_skb_hint = NULL;
1049 	tp->fastpath_skb_hint = NULL;
1050 }
1051 
1052 /* MD5 Signature */
1053 struct crypto_hash;
1054 
1055 /* - key database */
1056 struct tcp_md5sig_key {
1057 	u8			*key;
1058 	u8			keylen;
1059 };
1060 
1061 struct tcp4_md5sig_key {
1062 	u8			*key;
1063 	u16			keylen;
1064 	__be32			addr;
1065 };
1066 
1067 struct tcp6_md5sig_key {
1068 	u8			*key;
1069 	u16			keylen;
1070 #if 0
1071 	u32			scope_id;	/* XXX */
1072 #endif
1073 	struct in6_addr		addr;
1074 };
1075 
1076 /* - sock block */
1077 struct tcp_md5sig_info {
1078 	struct tcp4_md5sig_key	*keys4;
1079 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1080 	struct tcp6_md5sig_key	*keys6;
1081 	u32			entries6;
1082 	u32			alloced6;
1083 #endif
1084 	u32			entries4;
1085 	u32			alloced4;
1086 };
1087 
1088 /* - pseudo header */
1089 struct tcp4_pseudohdr {
1090 	__be32		saddr;
1091 	__be32		daddr;
1092 	__u8		pad;
1093 	__u8		protocol;
1094 	__be16		len;
1095 };
1096 
1097 struct tcp6_pseudohdr {
1098 	struct in6_addr	saddr;
1099 	struct in6_addr daddr;
1100 	__be32		len;
1101 	__be32		protocol;	/* including padding */
1102 };
1103 
1104 union tcp_md5sum_block {
1105 	struct tcp4_pseudohdr ip4;
1106 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1107 	struct tcp6_pseudohdr ip6;
1108 #endif
1109 };
1110 
1111 /* - pool: digest algorithm, hash description and scratch buffer */
1112 struct tcp_md5sig_pool {
1113 	struct hash_desc	md5_desc;
1114 	union tcp_md5sum_block	md5_blk;
1115 };
1116 
1117 #define TCP_MD5SIG_MAXKEYS	(~(u32)0)	/* really?! */
1118 
1119 /* - functions */
1120 extern int			tcp_v4_calc_md5_hash(char *md5_hash,
1121 						     struct tcp_md5sig_key *key,
1122 						     struct sock *sk,
1123 						     struct dst_entry *dst,
1124 						     struct request_sock *req,
1125 						     struct tcphdr *th,
1126 						     int protocol, int tcplen);
1127 extern struct tcp_md5sig_key	*tcp_v4_md5_lookup(struct sock *sk,
1128 						   struct sock *addr_sk);
1129 
1130 extern int			tcp_v4_md5_do_add(struct sock *sk,
1131 						  __be32 addr,
1132 						  u8 *newkey,
1133 						  u8 newkeylen);
1134 
1135 extern int			tcp_v4_md5_do_del(struct sock *sk,
1136 						  __be32 addr);
1137 
1138 extern struct tcp_md5sig_pool	**tcp_alloc_md5sig_pool(void);
1139 extern void			tcp_free_md5sig_pool(void);
1140 
1141 extern struct tcp_md5sig_pool	*__tcp_get_md5sig_pool(int cpu);
1142 extern void			__tcp_put_md5sig_pool(void);
1143 
1144 static inline
1145 struct tcp_md5sig_pool		*tcp_get_md5sig_pool(void)
1146 {
1147 	int cpu = get_cpu();
1148 	struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1149 	if (!ret)
1150 		put_cpu();
1151 	return ret;
1152 }
1153 
1154 static inline void		tcp_put_md5sig_pool(void)
1155 {
1156 	__tcp_put_md5sig_pool();
1157 	put_cpu();
1158 }
1159 
1160 /* write queue abstraction */
1161 static inline void tcp_write_queue_purge(struct sock *sk)
1162 {
1163 	struct sk_buff *skb;
1164 
1165 	while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1166 		sk_stream_free_skb(sk, skb);
1167 	sk_stream_mem_reclaim(sk);
1168 }
1169 
1170 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1171 {
1172 	struct sk_buff *skb = sk->sk_write_queue.next;
1173 	if (skb == (struct sk_buff *) &sk->sk_write_queue)
1174 		return NULL;
1175 	return skb;
1176 }
1177 
1178 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1179 {
1180 	struct sk_buff *skb = sk->sk_write_queue.prev;
1181 	if (skb == (struct sk_buff *) &sk->sk_write_queue)
1182 		return NULL;
1183 	return skb;
1184 }
1185 
1186 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1187 {
1188 	return skb->next;
1189 }
1190 
1191 #define tcp_for_write_queue(skb, sk)					\
1192 		for (skb = (sk)->sk_write_queue.next;			\
1193 		     (skb != (struct sk_buff *)&(sk)->sk_write_queue);	\
1194 		     skb = skb->next)
1195 
1196 #define tcp_for_write_queue_from(skb, sk)				\
1197 		for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1198 		     skb = skb->next)
1199 
1200 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1201 {
1202 	return sk->sk_send_head;
1203 }
1204 
1205 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1206 {
1207 	struct tcp_sock *tp = tcp_sk(sk);
1208 
1209 	sk->sk_send_head = skb->next;
1210 	if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1211 		sk->sk_send_head = NULL;
1212 	/* Don't override Nagle indefinately with F-RTO */
1213 	if (tp->frto_counter == 2)
1214 		tp->frto_counter = 3;
1215 }
1216 
1217 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1218 {
1219 	if (sk->sk_send_head == skb_unlinked)
1220 		sk->sk_send_head = NULL;
1221 }
1222 
1223 static inline void tcp_init_send_head(struct sock *sk)
1224 {
1225 	sk->sk_send_head = NULL;
1226 }
1227 
1228 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1229 {
1230 	__skb_queue_tail(&sk->sk_write_queue, skb);
1231 }
1232 
1233 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1234 {
1235 	__tcp_add_write_queue_tail(sk, skb);
1236 
1237 	/* Queue it, remembering where we must start sending. */
1238 	if (sk->sk_send_head == NULL)
1239 		sk->sk_send_head = skb;
1240 }
1241 
1242 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1243 {
1244 	__skb_queue_head(&sk->sk_write_queue, skb);
1245 }
1246 
1247 /* Insert buff after skb on the write queue of sk.  */
1248 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1249 						struct sk_buff *buff,
1250 						struct sock *sk)
1251 {
1252 	__skb_append(skb, buff, &sk->sk_write_queue);
1253 }
1254 
1255 /* Insert skb between prev and next on the write queue of sk.  */
1256 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1257 						  struct sk_buff *skb,
1258 						  struct sock *sk)
1259 {
1260 	__skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1261 }
1262 
1263 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1264 {
1265 	__skb_unlink(skb, &sk->sk_write_queue);
1266 }
1267 
1268 static inline int tcp_skb_is_last(const struct sock *sk,
1269 				  const struct sk_buff *skb)
1270 {
1271 	return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1272 }
1273 
1274 static inline int tcp_write_queue_empty(struct sock *sk)
1275 {
1276 	return skb_queue_empty(&sk->sk_write_queue);
1277 }
1278 
1279 /* /proc */
1280 enum tcp_seq_states {
1281 	TCP_SEQ_STATE_LISTENING,
1282 	TCP_SEQ_STATE_OPENREQ,
1283 	TCP_SEQ_STATE_ESTABLISHED,
1284 	TCP_SEQ_STATE_TIME_WAIT,
1285 };
1286 
1287 struct tcp_seq_afinfo {
1288 	struct module		*owner;
1289 	char			*name;
1290 	sa_family_t		family;
1291 	int			(*seq_show) (struct seq_file *m, void *v);
1292 	struct file_operations	*seq_fops;
1293 };
1294 
1295 struct tcp_iter_state {
1296 	sa_family_t		family;
1297 	enum tcp_seq_states	state;
1298 	struct sock		*syn_wait_sk;
1299 	int			bucket, sbucket, num, uid;
1300 	struct seq_operations	seq_ops;
1301 };
1302 
1303 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1304 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1305 
1306 extern struct request_sock_ops tcp_request_sock_ops;
1307 
1308 extern int tcp_v4_destroy_sock(struct sock *sk);
1309 
1310 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1311 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1312 
1313 #ifdef CONFIG_PROC_FS
1314 extern int  tcp4_proc_init(void);
1315 extern void tcp4_proc_exit(void);
1316 #endif
1317 
1318 /* TCP af-specific functions */
1319 struct tcp_sock_af_ops {
1320 #ifdef CONFIG_TCP_MD5SIG
1321 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1322 						struct sock *addr_sk);
1323 	int			(*calc_md5_hash) (char *location,
1324 						  struct tcp_md5sig_key *md5,
1325 						  struct sock *sk,
1326 						  struct dst_entry *dst,
1327 						  struct request_sock *req,
1328 						  struct tcphdr *th,
1329 						  int protocol, int len);
1330 	int			(*md5_add) (struct sock *sk,
1331 					    struct sock *addr_sk,
1332 					    u8 *newkey,
1333 					    u8 len);
1334 	int			(*md5_parse) (struct sock *sk,
1335 					      char __user *optval,
1336 					      int optlen);
1337 #endif
1338 };
1339 
1340 struct tcp_request_sock_ops {
1341 #ifdef CONFIG_TCP_MD5SIG
1342 	struct tcp_md5sig_key	*(*md5_lookup) (struct sock *sk,
1343 						struct request_sock *req);
1344 #endif
1345 };
1346 
1347 extern void tcp_v4_init(struct net_proto_family *ops);
1348 extern void tcp_init(void);
1349 
1350 #endif	/* _TCP_H */
1351