projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[NET]: Introduce inet_connection_sock
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / include / net / tcp.h
CommitLineData
1da177e4
LT		 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 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 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
463c84b9 22#define INET_CSK_DEBUG 1
1da177e4
LT		 23#define FASTRETRANS_DEBUG 1
24
25/* Cancel timers, when they are not required. */
463c84b9 26#undef INET_CSK_CLEAR_TIMERS
1da177e4
LT		 27
28#include <linux/config.h>
29#include <linux/list.h>
30#include <linux/tcp.h>
31#include <linux/slab.h>
32#include <linux/cache.h>
33#include <linux/percpu.h>
77d8bf9c 34#include <net/inet_hashtables.h>
1da177e4 35#include <net/checksum.h>
2e6599cb 36#include <net/request_sock.h>
1da177e4
LT		 37#include <net/sock.h>
38#include <net/snmp.h>
39#include <net/ip.h>
c752f073
ACM		 40#include <net/tcp_states.h>
41
1da177e4
LT		 42#include <linux/seq_file.h>
43
6e04e021 44extern struct inet_hashinfo tcp_hashinfo;
1da177e4 45
1da177e4
LT		 46extern atomic_t tcp_orphan_count;
47extern int tcp_tw_count;
48extern void tcp_time_wait(struct sock *sk, int state, int timeo);
8feaf0c0 49extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
1da177e4 50
1da177e4
LT		 51#define MAX_TCP_HEADER (128 + MAX_HEADER)
52
53/*
54 * Never offer a window over 32767 without using window scaling. Some
55 * poor stacks do signed 16bit maths!
56 */
57#define MAX_TCP_WINDOW 32767U
58
59/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
60#define TCP_MIN_MSS 88U
61
62/* Minimal RCV_MSS. */
63#define TCP_MIN_RCVMSS 536U
64
65/* After receiving this amount of duplicate ACKs fast retransmit starts. */
66#define TCP_FASTRETRANS_THRESH 3
67
68/* Maximal reordering. */
69#define TCP_MAX_REORDERING 127
70
71/* Maximal number of ACKs sent quickly to accelerate slow-start. */
72#define TCP_MAX_QUICKACKS 16U
73
74/* urg_data states */
75#define TCP_URG_VALID 0x0100
76#define TCP_URG_NOTYET 0x0200
77#define TCP_URG_READ 0x0400
78
79#define TCP_RETR1 3 /*
80 * This is how many retries it does before it
81 * tries to figure out if the gateway is
82 * down. Minimal RFC value is 3; it corresponds
83 * to ~3sec-8min depending on RTO.
84 */
85
86#define TCP_RETR2 15 /*
87 * This should take at least
88 * 90 minutes to time out.
89 * RFC1122 says that the limit is 100 sec.
90 * 15 is ~13-30min depending on RTO.
91 */
92
93#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
94 * connection: ~180sec is RFC minumum */
95
96#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
97 * connection: ~180sec is RFC minumum */
98
99
100#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
101 * socket. 7 is ~50sec-16min.
102 */
103
104
105#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
106 * state, about 60 seconds */
107#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
108 /* BSD style FIN_WAIT2 deadlock breaker.
109 * It used to be 3min, new value is 60sec,
110 * to combine FIN-WAIT-2 timeout with
111 * TIME-WAIT timer.
112 */
113
114#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
115#if HZ >= 100
116#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
117#define TCP_ATO_MIN ((unsigned)(HZ/25))
118#else
119#define TCP_DELACK_MIN 4U
120#define TCP_ATO_MIN 4U
121#endif
122#define TCP_RTO_MAX ((unsigned)(120*HZ))
123#define TCP_RTO_MIN ((unsigned)(HZ/5))
124#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
125
126#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
127 * for local resources.
128 */
129
130#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
131#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
132#define TCP_KEEPALIVE_INTVL (75*HZ)
133
134#define MAX_TCP_KEEPIDLE 32767
135#define MAX_TCP_KEEPINTVL 32767
136#define MAX_TCP_KEEPCNT 127
137#define MAX_TCP_SYNCNT 127
138
139#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
140#define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
141
142#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
143#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
144 * after this time. It should be equal
145 * (or greater than) TCP_TIMEWAIT_LEN
146 * to provide reliability equal to one
147 * provided by timewait state.
148 */
149#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
150 * timestamps. It must be less than
151 * minimal timewait lifetime.
152 */
153
154#define TCP_TW_RECYCLE_SLOTS_LOG 5
155#define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
156
157/* If time > 4sec, it is "slow" path, no recycling is required,
158 so that we select tick to get range about 4 seconds.
159 */
160
161#if HZ <= 16 || HZ > 4096
162# error Unsupported: HZ <= 16 or HZ > 4096
163#elif HZ <= 32
164# define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
165#elif HZ <= 64
166# define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
167#elif HZ <= 128
168# define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
169#elif HZ <= 256
170# define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
171#elif HZ <= 512
172# define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
173#elif HZ <= 1024
174# define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
175#elif HZ <= 2048
176# define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
177#else
178# define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
179#endif
1da177e4
LT		 180/*
181 * TCP option
182 */
183
184#define TCPOPT_NOP 1 /* Padding */
185#define TCPOPT_EOL 0 /* End of options */
186#define TCPOPT_MSS 2 /* Segment size negotiating */
187#define TCPOPT_WINDOW 3 /* Window scaling */
188#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
189#define TCPOPT_SACK 5 /* SACK Block */
190#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
191
192/*
193 * TCP option lengths
194 */
195
196#define TCPOLEN_MSS 4
197#define TCPOLEN_WINDOW 3
198#define TCPOLEN_SACK_PERM 2
199#define TCPOLEN_TIMESTAMP 10
200
201/* But this is what stacks really send out. */
202#define TCPOLEN_TSTAMP_ALIGNED 12
203#define TCPOLEN_WSCALE_ALIGNED 4
204#define TCPOLEN_SACKPERM_ALIGNED 4
205#define TCPOLEN_SACK_BASE 2
206#define TCPOLEN_SACK_BASE_ALIGNED 4
207#define TCPOLEN_SACK_PERBLOCK 8
208
463c84b9
ACM		 209#define ICSK_TIME_RETRANS 1 /* Retransmit timer */
210#define ICSK_TIME_DACK 2 /* Delayed ack timer */
211#define ICSK_TIME_PROBE0 3 /* Zero window probe timer */
212#define ICSK_TIME_KEEPOPEN 4 /* Keepalive timer */
1da177e4
LT		 213
214/* Flags in tp->nonagle */
215#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
216#define TCP_NAGLE_CORK 2 /* Socket is corked */
217#define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
218
219/* sysctl variables for tcp */
1da177e4
LT		 220extern int sysctl_tcp_timestamps;
221extern int sysctl_tcp_window_scaling;
222extern int sysctl_tcp_sack;
223extern int sysctl_tcp_fin_timeout;
224extern int sysctl_tcp_tw_recycle;
225extern int sysctl_tcp_keepalive_time;
226extern int sysctl_tcp_keepalive_probes;
227extern int sysctl_tcp_keepalive_intvl;
228extern int sysctl_tcp_syn_retries;
229extern int sysctl_tcp_synack_retries;
230extern int sysctl_tcp_retries1;
231extern int sysctl_tcp_retries2;
232extern int sysctl_tcp_orphan_retries;
233extern int sysctl_tcp_syncookies;
234extern int sysctl_tcp_retrans_collapse;
235extern int sysctl_tcp_stdurg;
236extern int sysctl_tcp_rfc1337;
237extern int sysctl_tcp_abort_on_overflow;
238extern int sysctl_tcp_max_orphans;
239extern int sysctl_tcp_max_tw_buckets;
240extern int sysctl_tcp_fack;
241extern int sysctl_tcp_reordering;
242extern int sysctl_tcp_ecn;
243extern int sysctl_tcp_dsack;
244extern int sysctl_tcp_mem[3];
245extern int sysctl_tcp_wmem[3];
246extern int sysctl_tcp_rmem[3];
247extern int sysctl_tcp_app_win;
248extern int sysctl_tcp_adv_win_scale;
249extern int sysctl_tcp_tw_reuse;
250extern int sysctl_tcp_frto;
251extern int sysctl_tcp_low_latency;
1da177e4 252extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 253extern int sysctl_tcp_moderate_rcvbuf;
254extern int sysctl_tcp_tso_win_divisor;
255
256extern atomic_t tcp_memory_allocated;
257extern atomic_t tcp_sockets_allocated;
258extern int tcp_memory_pressure;
259
1da177e4 260#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
463c84b9 261#define AF_INET_FAMILY(fam) ((fam) == AF_INET)
1da177e4 262#else
463c84b9 263#define AF_INET_FAMILY(fam) 1
1da177e4
LT		 264#endif
265
266/*
267 * Pointers to address related TCP functions
268 * (i.e. things that depend on the address family)
269 */
270
271struct tcp_func {
272 int (*queue_xmit) (struct sk_buff *skb,
273 int ipfragok);
274
275 void (*send_check) (struct sock *sk,
276 struct tcphdr *th,
277 int len,
278 struct sk_buff *skb);
279
280 int (*rebuild_header) (struct sock *sk);
281
282 int (*conn_request) (struct sock *sk,
283 struct sk_buff *skb);
284
285 struct sock * (*syn_recv_sock) (struct sock *sk,
286 struct sk_buff *skb,
60236fdd 287 struct request_sock *req,
1da177e4
LT		 288 struct dst_entry *dst);
289
290 int (*remember_stamp) (struct sock *sk);
291
292 __u16 net_header_len;
293
294 int (*setsockopt) (struct sock *sk,
295 int level,
296 int optname,
297 char __user *optval,
298 int optlen);
299
300 int (*getsockopt) (struct sock *sk,
301 int level,
302 int optname,
303 char __user *optval,
304 int __user *optlen);
305
306
307 void (*addr2sockaddr) (struct sock *sk,
308 struct sockaddr *);
309
310 int sockaddr_len;
311};
312
313/*
314 * The next routines deal with comparing 32 bit unsigned ints
315 * and worry about wraparound (automatic with unsigned arithmetic).
316 */
317
318static inline int before(__u32 seq1, __u32 seq2)
319{
320 return (__s32)(seq1-seq2) < 0;
321}
322
323static inline int after(__u32 seq1, __u32 seq2)
324{
325 return (__s32)(seq2-seq1) < 0;
326}
327
328
329/* is s2<=s1<=s3 ? */
330static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
331{
332 return seq3 - seq2 >= seq1 - seq2;
333}
334
335
336extern struct proto tcp_prot;
337
338DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
339#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
340#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
341#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
342#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
343#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
344#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
345
1da177e4
LT		 346extern void tcp_v4_err(struct sk_buff *skb, u32);
347
348extern void tcp_shutdown (struct sock *sk, int how);
349
350extern int tcp_v4_rcv(struct sk_buff *skb);
351
352extern int tcp_v4_remember_stamp(struct sock *sk);
353
8feaf0c0 354extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4
LT		 355
356extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
357 struct msghdr *msg, size_t size);
358extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
359
360extern int tcp_ioctl(struct sock *sk,
361 int cmd,
362 unsigned long arg);
363
364extern int tcp_rcv_state_process(struct sock *sk,
365 struct sk_buff *skb,
366 struct tcphdr *th,
367 unsigned len);
368
369extern int tcp_rcv_established(struct sock *sk,
370 struct sk_buff *skb,
371 struct tcphdr *th,
372 unsigned len);
373
374extern void tcp_rcv_space_adjust(struct sock *sk);
375
463c84b9
ACM		 376enum inet_csk_ack_state_t {
377 ICSK_ACK_SCHED = 1,
378 ICSK_ACK_TIMER = 2,
379 ICSK_ACK_PUSHED = 4
1da177e4
LT		 380};
381
463c84b9 382static inline void inet_csk_schedule_ack(struct sock *sk)
1da177e4 383{
463c84b9 384 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_SCHED;
1da177e4
LT		 385}
386
463c84b9 387static inline int inet_csk_ack_scheduled(const struct sock *sk)
1da177e4 388{
463c84b9 389 return inet_csk(sk)->icsk_ack.pending & ICSK_ACK_SCHED;
1da177e4
LT		 390}
391
463c84b9
ACM		 392static inline void tcp_dec_quickack_mode(struct sock *sk,
393 const unsigned int pkts)
1da177e4 394{
463c84b9 395 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 396
463c84b9
ACM		 397 if (icsk->icsk_ack.quick) {
398 if (pkts >= icsk->icsk_ack.quick) {
399 icsk->icsk_ack.quick = 0;
fc6415bc 400 /* Leaving quickack mode we deflate ATO. */
463c84b9 401 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 402 } else
463c84b9 403 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 404 }
405}
406
463c84b9 407extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 408
463c84b9 409static inline void inet_csk_delack_init(struct sock *sk)
1da177e4 410{
463c84b9 411 memset(&inet_csk(sk)->icsk_ack, 0, sizeof(inet_csk(sk)->icsk_ack));
1da177e4
LT		 412}
413
414static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
415{
416 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
417}
418
419enum tcp_tw_status
420{
421 TCP_TW_SUCCESS = 0,
422 TCP_TW_RST = 1,
423 TCP_TW_ACK = 2,
424 TCP_TW_SYN = 3
425};
426
427
8feaf0c0 428extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 429 struct sk_buff *skb,
8feaf0c0 430 const struct tcphdr *th);
1da177e4
LT		 431
432extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 433 struct request_sock *req,
434 struct request_sock **prev);
1da177e4
LT		 435extern int tcp_child_process(struct sock *parent,
436 struct sock *child,
437 struct sk_buff *skb);
438extern void tcp_enter_frto(struct sock *sk);
439extern void tcp_enter_loss(struct sock *sk, int how);
440extern void tcp_clear_retrans(struct tcp_sock *tp);
441extern void tcp_update_metrics(struct sock *sk);
442
443extern void tcp_close(struct sock *sk,
444 long timeout);
463c84b9 445extern struct sock * inet_csk_accept(struct sock *sk, int flags, int *err);
1da177e4
LT		 446extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
447
448extern int tcp_getsockopt(struct sock *sk, int level,
449 int optname,
450 char __user *optval,
451 int __user *optlen);
452extern int tcp_setsockopt(struct sock *sk, int level,
453 int optname, char __user *optval,
454 int optlen);
455extern void tcp_set_keepalive(struct sock *sk, int val);
456extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
457 struct msghdr *msg,
458 size_t len, int nonblock,
459 int flags, int *addr_len);
460
461extern int tcp_listen_start(struct sock *sk);
462
463extern void tcp_parse_options(struct sk_buff *skb,
464 struct tcp_options_received *opt_rx,
465 int estab);
466
467/*
468 * TCP v4 functions exported for the inet6 API
469 */
470
1da177e4
LT		 471extern void tcp_v4_send_check(struct sock *sk,
472 struct tcphdr *th, int len,
473 struct sk_buff *skb);
474
475extern int tcp_v4_conn_request(struct sock *sk,
476 struct sk_buff *skb);
477
478extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 479 struct request_sock *req,
1da177e4
LT		 480 struct sk_buff *skb);
481
482extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
483 struct sk_buff *skb,
60236fdd 484 struct request_sock *req,
1da177e4
LT		 485 struct dst_entry *dst);
486
487extern int tcp_v4_do_rcv(struct sock *sk,
488 struct sk_buff *skb);
489
490extern int tcp_v4_connect(struct sock *sk,
491 struct sockaddr *uaddr,
492 int addr_len);
493
494extern int tcp_connect(struct sock *sk);
495
496extern struct sk_buff * tcp_make_synack(struct sock *sk,
497 struct dst_entry *dst,
60236fdd 498 struct request_sock *req);
1da177e4
LT		 499
500extern int tcp_disconnect(struct sock *sk, int flags);
501
502extern void tcp_unhash(struct sock *sk);
503
504extern int tcp_v4_hash_connecting(struct sock *sk);
505
506
507/* From syncookies.c */
508extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
509 struct ip_options *opt);
510extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
511 __u16 *mss);
512
513/* tcp_output.c */
514
f6302d1d 515extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
a2e2a59c 516 unsigned int cur_mss, int nonagle);
f6302d1d 517extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
1da177e4
LT		 518extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
519extern void tcp_xmit_retransmit_queue(struct sock *);
520extern void tcp_simple_retransmit(struct sock *);
521extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
522
523extern void tcp_send_probe0(struct sock *);
524extern void tcp_send_partial(struct sock *);
525extern int tcp_write_wakeup(struct sock *);
526extern void tcp_send_fin(struct sock *sk);
86a76caf
VF		 527extern void tcp_send_active_reset(struct sock *sk,
528 unsigned int __nocast priority);
1da177e4 529extern int tcp_send_synack(struct sock *);
c1b4a7e6 530extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 531extern void tcp_send_ack(struct sock *sk);
532extern void tcp_send_delayed_ack(struct sock *sk);
533
a762a980
DM		 534/* tcp_input.c */
535extern void tcp_cwnd_application_limited(struct sock *sk);
536
1da177e4
LT		 537/* tcp_timer.c */
538extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 539static inline void tcp_clear_xmit_timers(struct sock *sk)
540{
541 inet_csk_clear_xmit_timers(sk);
542}
1da177e4 543
463c84b9
ACM		 544extern void inet_csk_delete_keepalive_timer(struct sock *sk);
545extern void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long timeout);
1da177e4
LT		 546extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
547extern unsigned int tcp_current_mss(struct sock *sk, int large);
548
463c84b9
ACM		 549#ifdef INET_CSK_DEBUG
550extern const char inet_csk_timer_bug_msg[];
1da177e4
LT		 551#endif
552
553/* tcp_diag.c */
554extern void tcp_get_info(struct sock *, struct tcp_info *);
555
556/* Read 'sendfile()'-style from a TCP socket */
557typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
558 unsigned int, size_t);
559extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
560 sk_read_actor_t recv_actor);
561
463c84b9 562static inline void inet_csk_clear_xmit_timer(struct sock *sk, const int what)
1da177e4 563{
463c84b9 564 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 565
463c84b9
ACM		 566 if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
567 icsk->icsk_pending = 0;
568#ifdef INET_CSK_CLEAR_TIMERS
569 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
1da177e4 570#endif
463c84b9
ACM		 571 } else if (what == ICSK_TIME_DACK) {
572 icsk->icsk_ack.blocked = icsk->icsk_ack.pending = 0;
573#ifdef INET_CSK_CLEAR_TIMERS
574 sk_stop_timer(sk, &icsk->icsk_delack_timer);
1da177e4 575#endif
463c84b9
ACM		 576 }
577#ifdef INET_CSK_DEBUG
578 else {
579 pr_debug(inet_csk_timer_bug_msg);
580 }
1da177e4 581#endif
1da177e4
LT		 582}
583
584/*
585 * Reset the retransmission timer
586 */
463c84b9
ACM		 587static inline void inet_csk_reset_xmit_timer(struct sock *sk, const int what,
588 unsigned long when)
1da177e4 589{
463c84b9 590 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT		 591
592 if (when > TCP_RTO_MAX) {
463c84b9
ACM		 593#ifdef INET_CSK_DEBUG
594 pr_debug("reset_xmit_timer: sk=%p %d when=0x%lx, caller=%p\n",
595 sk, what, when, current_text_addr());
1da177e4
LT		 596#endif
597 when = TCP_RTO_MAX;
598 }
599
463c84b9
ACM		 600 if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
601 icsk->icsk_pending = what;
602 icsk->icsk_timeout = jiffies + when;
603 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
604 } else if (what == ICSK_TIME_DACK) {
605 icsk->icsk_ack.pending |= ICSK_ACK_TIMER;
606 icsk->icsk_ack.timeout = jiffies + when;
607 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
608 }
609#ifdef INET_CSK_DEBUG
610 else {
611 pr_debug(inet_csk_timer_bug_msg);
612 }
1da177e4 613#endif
1da177e4
LT		 614}
615
616/* Initialize RCV_MSS value.
617 * RCV_MSS is an our guess about MSS used by the peer.
618 * We haven't any direct information about the MSS.
619 * It's better to underestimate the RCV_MSS rather than overestimate.
620 * Overestimations make us ACKing less frequently than needed.
621 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
622 */
623
624static inline void tcp_initialize_rcv_mss(struct sock *sk)
625{
626 struct tcp_sock *tp = tcp_sk(sk);
c1b4a7e6 627 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
1da177e4
LT		 628
629 hint = min(hint, tp->rcv_wnd/2);
630 hint = min(hint, TCP_MIN_RCVMSS);
631 hint = max(hint, TCP_MIN_MSS);
632
463c84b9 633 inet_csk(sk)->icsk_ack.rcv_mss = hint;
1da177e4
LT		 634}
635
636static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
637{
638 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
639 ntohl(TCP_FLAG_ACK) |
640 snd_wnd);
641}
642
643static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
644{
645 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
646}
647
648static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
649{
b03efcfb 650 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 651 tp->rcv_wnd &&
652 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
653 !tp->urg_data)
654 tcp_fast_path_on(tp);
655}
656
657/* Compute the actual receive window we are currently advertising.
658 * Rcv_nxt can be after the window if our peer push more data
659 * than the offered window.
660 */
661static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
662{
663 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
664
665 if (win < 0)
666 win = 0;
667 return (u32) win;
668}
669
670/* Choose a new window, without checks for shrinking, and without
671 * scaling applied to the result. The caller does these things
672 * if necessary. This is a "raw" window selection.
673 */
674extern u32 __tcp_select_window(struct sock *sk);
675
676/* TCP timestamps are only 32-bits, this causes a slight
677 * complication on 64-bit systems since we store a snapshot
678 * of jiffies in the buffer control blocks below. We decidely
679 * only use of the low 32-bits of jiffies and hide the ugly
680 * casts with the following macro.
681 */
682#define tcp_time_stamp ((__u32)(jiffies))
683
684/* This is what the send packet queueing engine uses to pass
685 * TCP per-packet control information to the transmission
686 * code. We also store the host-order sequence numbers in
687 * here too. This is 36 bytes on 32-bit architectures,
688 * 40 bytes on 64-bit machines, if this grows please adjust
689 * skbuff.h:skbuff->cb[xxx] size appropriately.
690 */
691struct tcp_skb_cb {
692 union {
693 struct inet_skb_parm h4;
694#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
695 struct inet6_skb_parm h6;
696#endif
697 } header; /* For incoming frames */
698 __u32 seq; /* Starting sequence number */
699 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
700 __u32 when; /* used to compute rtt's */
701 __u8 flags; /* TCP header flags. */
702
703 /* NOTE: These must match up to the flags byte in a
704 * real TCP header.
705 */
706#define TCPCB_FLAG_FIN 0x01
707#define TCPCB_FLAG_SYN 0x02
708#define TCPCB_FLAG_RST 0x04
709#define TCPCB_FLAG_PSH 0x08
710#define TCPCB_FLAG_ACK 0x10
711#define TCPCB_FLAG_URG 0x20
712#define TCPCB_FLAG_ECE 0x40
713#define TCPCB_FLAG_CWR 0x80
714
715 __u8 sacked; /* State flags for SACK/FACK. */
716#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
717#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
718#define TCPCB_LOST 0x04 /* SKB is lost */
719#define TCPCB_TAGBITS 0x07 /* All tag bits */
720
721#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
722#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
723
724#define TCPCB_URG 0x20 /* Urgent pointer advenced here */
725
726#define TCPCB_AT_TAIL (TCPCB_URG)
727
728 __u16 urg_ptr; /* Valid w/URG flags is set. */
729 __u32 ack_seq; /* Sequence number ACK'd */
730};
731
732#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
733
734#include <net/tcp_ecn.h>
735
736/* Due to TSO, an SKB can be composed of multiple actual
737 * packets. To keep these tracked properly, we use this.
738 */
739static inline int tcp_skb_pcount(const struct sk_buff *skb)
740{
741 return skb_shinfo(skb)->tso_segs;
742}
743
744/* This is valid iff tcp_skb_pcount() > 1. */
745static inline int tcp_skb_mss(const struct sk_buff *skb)
746{
747 return skb_shinfo(skb)->tso_size;
748}
749
750static inline void tcp_dec_pcount_approx(__u32 *count,
751 const struct sk_buff *skb)
752{
753 if (*count) {
754 *count -= tcp_skb_pcount(skb);
755 if ((int)*count < 0)
756 *count = 0;
757 }
758}
759
760static inline void tcp_packets_out_inc(struct sock *sk,
761 struct tcp_sock *tp,
762 const struct sk_buff *skb)
763{
764 int orig = tp->packets_out;
765
766 tp->packets_out += tcp_skb_pcount(skb);
767 if (!orig)
463c84b9 768 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, inet_csk(sk)->icsk_rto);
1da177e4
LT		 769}
770
771static inline void tcp_packets_out_dec(struct tcp_sock *tp,
772 const struct sk_buff *skb)
773{
774 tp->packets_out -= tcp_skb_pcount(skb);
775}
776
317a76f9
SH		 777/* Events passed to congestion control interface */
778enum tcp_ca_event {
779 CA_EVENT_TX_START, /* first transmit when no packets in flight */
780 CA_EVENT_CWND_RESTART, /* congestion window restart */
781 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
782 CA_EVENT_FRTO, /* fast recovery timeout */
783 CA_EVENT_LOSS, /* loss timeout */
784 CA_EVENT_FAST_ACK, /* in sequence ack */
785 CA_EVENT_SLOW_ACK, /* other ack */
786};
787
788/*
789 * Interface for adding new TCP congestion control handlers
790 */
791#define TCP_CA_NAME_MAX 16
792struct tcp_congestion_ops {
793 struct list_head list;
794
795 /* initialize private data (optional) */
796 void (*init)(struct tcp_sock *tp);
797 /* cleanup private data (optional) */
798 void (*release)(struct tcp_sock *tp);
799
800 /* return slow start threshold (required) */
801 u32 (*ssthresh)(struct tcp_sock *tp);
802 /* lower bound for congestion window (optional) */
803 u32 (*min_cwnd)(struct tcp_sock *tp);
804 /* do new cwnd calculation (required) */
805 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
806 u32 rtt, u32 in_flight, int good_ack);
807 /* round trip time sample per acked packet (optional) */
808 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
809 /* call before changing ca_state (optional) */
810 void (*set_state)(struct tcp_sock *tp, u8 new_state);
811 /* call when cwnd event occurs (optional) */
812 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
813 /* new value of cwnd after loss (optional) */
814 u32 (*undo_cwnd)(struct tcp_sock *tp);
815 /* hook for packet ack accounting (optional) */
816 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
817 /* get info for tcp_diag (optional) */
818 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
819
820 char name[TCP_CA_NAME_MAX];
821 struct module *owner;
822};
823
824extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
825extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
826
827extern void tcp_init_congestion_control(struct tcp_sock *tp);
828extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
829extern int tcp_set_default_congestion_control(const char *name);
830extern void tcp_get_default_congestion_control(char *name);
5f8ef48d 831extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
317a76f9 832
5f8ef48d 833extern struct tcp_congestion_ops tcp_init_congestion_ops;
317a76f9
SH		 834extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
835extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
836 u32 rtt, u32 in_flight, int flag);
837extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
a8acfbac 838extern struct tcp_congestion_ops tcp_reno;
317a76f9
SH		 839
840static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
841{
842 if (tp->ca_ops->set_state)
843 tp->ca_ops->set_state(tp, ca_state);
844 tp->ca_state = ca_state;
845}
846
847static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
848{
849 if (tp->ca_ops->cwnd_event)
850 tp->ca_ops->cwnd_event(tp, event);
851}
852
1da177e4
LT		 853/* This determines how many packets are "in the network" to the best
854 * of our knowledge. In many cases it is conservative, but where
855 * detailed information is available from the receiver (via SACK
856 * blocks etc.) we can make more aggressive calculations.
857 *
858 * Use this for decisions involving congestion control, use just
859 * tp->packets_out to determine if the send queue is empty or not.
860 *
861 * Read this equation as:
862 *
863 * "Packets sent once on transmission queue" MINUS
864 * "Packets left network, but not honestly ACKed yet" PLUS
865 * "Packets fast retransmitted"
866 */
867static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
868{
869 return (tp->packets_out - tp->left_out + tp->retrans_out);
870}
871
1da177e4
LT		 872/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
873 * The exception is rate halving phase, when cwnd is decreasing towards
874 * ssthresh.
875 */
876static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
877{
878 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
879 return tp->snd_ssthresh;
880 else
881 return max(tp->snd_ssthresh,
882 ((tp->snd_cwnd >> 1) +
883 (tp->snd_cwnd >> 2)));
884}
885
886static inline void tcp_sync_left_out(struct tcp_sock *tp)
887{
888 if (tp->rx_opt.sack_ok &&
889 (tp->sacked_out >= tp->packets_out - tp->lost_out))
890 tp->sacked_out = tp->packets_out - tp->lost_out;
891 tp->left_out = tp->sacked_out + tp->lost_out;
892}
893
d1b04c08 894/* Set slow start threshold and cwnd not falling to slow start */
1da177e4
LT		 895static inline void __tcp_enter_cwr(struct tcp_sock *tp)
896{
897 tp->undo_marker = 0;
317a76f9 898 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
1da177e4
LT		 899 tp->snd_cwnd = min(tp->snd_cwnd,
900 tcp_packets_in_flight(tp) + 1U);
901 tp->snd_cwnd_cnt = 0;
902 tp->high_seq = tp->snd_nxt;
903 tp->snd_cwnd_stamp = tcp_time_stamp;
904 TCP_ECN_queue_cwr(tp);
905}
906
907static inline void tcp_enter_cwr(struct tcp_sock *tp)
908{
909 tp->prior_ssthresh = 0;
910 if (tp->ca_state < TCP_CA_CWR) {
911 __tcp_enter_cwr(tp);
912 tcp_set_ca_state(tp, TCP_CA_CWR);
913 }
914}
915
916extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
917
918/* Slow start with delack produces 3 packets of burst, so that
919 * it is safe "de facto".
920 */
921static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
922{
923 return 3;
924}
925
1da177e4
LT		 926static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
927 const struct sk_buff *skb)
928{
929 if (skb->len < mss)
930 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
931}
932
1da177e4
LT		 933static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
934{
463c84b9
ACM		 935 const struct inet_connection_sock *icsk = inet_csk(sk);
936 if (!tp->packets_out && !icsk->icsk_pending)
937 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, icsk->icsk_rto);
1da177e4
LT		 938}
939
1da177e4
LT		 940static __inline__ void tcp_push_pending_frames(struct sock *sk,
941 struct tcp_sock *tp)
942{
943 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
944}
945
1da177e4
LT		 946static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
947{
948 tp->snd_wl1 = seq;
949}
950
951static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
952{
953 tp->snd_wl1 = seq;
954}
955
956extern void tcp_destroy_sock(struct sock *sk);
957
958
959/*
960 * Calculate(/check) TCP checksum
961 */
962static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
963 unsigned long saddr, unsigned long daddr,
964 unsigned long base)
965{
966 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
967}
968
969static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
970{
971 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
972}
973
974static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
975{
976 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
977 __tcp_checksum_complete(skb);
978}
979
980/* Prequeue for VJ style copy to user, combined with checksumming. */
981
982static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
983{
984 tp->ucopy.task = NULL;
985 tp->ucopy.len = 0;
986 tp->ucopy.memory = 0;
987 skb_queue_head_init(&tp->ucopy.prequeue);
988}
989
990/* Packet is added to VJ-style prequeue for processing in process
991 * context, if a reader task is waiting. Apparently, this exciting
992 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
993 * failed somewhere. Latency? Burstiness? Well, at least now we will
994 * see, why it failed. 8)8) --ANK
995 *
996 * NOTE: is this not too big to inline?
997 */
998static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
999{
1000 struct tcp_sock *tp = tcp_sk(sk);
1001
1002 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1003 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1004 tp->ucopy.memory += skb->truesize;
1005 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1006 struct sk_buff *skb1;
1007
1008 BUG_ON(sock_owned_by_user(sk));
1009
1010 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1011 sk->sk_backlog_rcv(sk, skb1);
1012 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1013 }
1014
1015 tp->ucopy.memory = 0;
1016 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1017 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 1018 if (!inet_csk_ack_scheduled(sk))
1019 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1020 (3 * TCP_RTO_MIN) / 4);
1da177e4
LT		 1021 }
1022 return 1;
1023 }
1024 return 0;
1025}
1026
1027
1028#undef STATE_TRACE
1029
1030#ifdef STATE_TRACE
1031static const char *statename[]={
1032 "Unused","Established","Syn Sent","Syn Recv",
1033 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1034 "Close Wait","Last ACK","Listen","Closing"
1035};
1036#endif
1037
1038static __inline__ void tcp_set_state(struct sock *sk, int state)
1039{
1040 int oldstate = sk->sk_state;
1041
1042 switch (state) {
1043 case TCP_ESTABLISHED:
1044 if (oldstate != TCP_ESTABLISHED)
1045 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1046 break;
1047
1048 case TCP_CLOSE:
1049 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1050 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1051
1052 sk->sk_prot->unhash(sk);
463c84b9 1053 if (inet_csk(sk)->icsk_bind_hash &&
1da177e4 1054 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2d8c4ce5 1055 inet_put_port(&tcp_hashinfo, sk);
1da177e4
LT		 1056 /* fall through */
1057 default:
1058 if (oldstate==TCP_ESTABLISHED)
1059 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1060 }
1061
1062 /* Change state AFTER socket is unhashed to avoid closed
1063 * socket sitting in hash tables.
1064 */
1065 sk->sk_state = state;
1066
1067#ifdef STATE_TRACE
1068 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1069#endif
1070}
1071
1072static __inline__ void tcp_done(struct sock *sk)
1073{
1074 tcp_set_state(sk, TCP_CLOSE);
1075 tcp_clear_xmit_timers(sk);
1076
1077 sk->sk_shutdown = SHUTDOWN_MASK;
1078
1079 if (!sock_flag(sk, SOCK_DEAD))
1080 sk->sk_state_change(sk);
1081 else
1082 tcp_destroy_sock(sk);
1083}
1084
1085static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1086{
1087 rx_opt->dsack = 0;
1088 rx_opt->eff_sacks = 0;
1089 rx_opt->num_sacks = 0;
1090}
1091
1092static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1093{
1094 if (tp->rx_opt.tstamp_ok) {
1095 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1096 (TCPOPT_NOP << 16) |
1097 (TCPOPT_TIMESTAMP << 8) |
1098 TCPOLEN_TIMESTAMP);
1099 *ptr++ = htonl(tstamp);
1100 *ptr++ = htonl(tp->rx_opt.ts_recent);
1101 }
1102 if (tp->rx_opt.eff_sacks) {
1103 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1104 int this_sack;
1105
1106 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1107 (TCPOPT_NOP << 16) |
1108 (TCPOPT_SACK << 8) |
1109 (TCPOLEN_SACK_BASE +
1110 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1111 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1112 *ptr++ = htonl(sp[this_sack].start_seq);
1113 *ptr++ = htonl(sp[this_sack].end_seq);
1114 }
1115 if (tp->rx_opt.dsack) {
1116 tp->rx_opt.dsack = 0;
1117 tp->rx_opt.eff_sacks--;
1118 }
1119 }
1120}
1121
1122/* Construct a tcp options header for a SYN or SYN_ACK packet.
1123 * If this is every changed make sure to change the definition of
1124 * MAX_SYN_SIZE to match the new maximum number of options that you
1125 * can generate.
1126 */
1127static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1128 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1129{
1130 /* We always get an MSS option.
1131 * The option bytes which will be seen in normal data
1132 * packets should timestamps be used, must be in the MSS
1133 * advertised. But we subtract them from tp->mss_cache so
1134 * that calculations in tcp_sendmsg are simpler etc.
1135 * So account for this fact here if necessary. If we
1136 * don't do this correctly, as a receiver we won't
1137 * recognize data packets as being full sized when we
1138 * should, and thus we won't abide by the delayed ACK
1139 * rules correctly.
1140 * SACKs don't matter, we never delay an ACK when we
1141 * have any of those going out.
1142 */
1143 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1144 if (ts) {
1145 if(sack)
1146 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1147 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1148 else
1149 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1150 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1151 *ptr++ = htonl(tstamp); /* TSVAL */
1152 *ptr++ = htonl(ts_recent); /* TSECR */
1153 } else if(sack)
1154 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1155 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1156 if (offer_wscale)
1157 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1158}
1159
1160/* Determine a window scaling and initial window to offer. */
1161extern void tcp_select_initial_window(int __space, __u32 mss,
1162 __u32 *rcv_wnd, __u32 *window_clamp,
1163 int wscale_ok, __u8 *rcv_wscale);
1164
1165static inline int tcp_win_from_space(int space)
1166{
1167 return sysctl_tcp_adv_win_scale<=0 ?
1168 (space>>(-sysctl_tcp_adv_win_scale)) :
1169 space - (space>>sysctl_tcp_adv_win_scale);
1170}
1171
1172/* Note: caller must be prepared to deal with negative returns */
1173static inline int tcp_space(const struct sock *sk)
1174{
1175 return tcp_win_from_space(sk->sk_rcvbuf -
1176 atomic_read(&sk->sk_rmem_alloc));
1177}
1178
1179static inline int tcp_full_space(const struct sock *sk)
1180{
1181 return tcp_win_from_space(sk->sk_rcvbuf);
1182}
1183
463c84b9
ACM		 1184static inline void inet_csk_reqsk_queue_add(struct sock *sk,
1185 struct request_sock *req,
1186 struct sock *child)
1da177e4 1187{
463c84b9 1188 reqsk_queue_add(&inet_csk(sk)->icsk_accept_queue, req, sk, child);
1da177e4
LT		 1189}
1190
463c84b9
ACM		 1191static inline void inet_csk_reqsk_queue_removed(struct sock *sk,
1192 struct request_sock *req)
1da177e4 1193{
463c84b9
ACM		 1194 if (reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req) == 0)
1195 inet_csk_delete_keepalive_timer(sk);
1da177e4
LT		 1196}
1197
463c84b9
ACM		 1198static inline void inet_csk_reqsk_queue_added(struct sock *sk,
1199 const unsigned long timeout)
1da177e4 1200{
463c84b9
ACM		 1201 if (reqsk_queue_added(&inet_csk(sk)->icsk_accept_queue) == 0)
1202 inet_csk_reset_keepalive_timer(sk, timeout);
1da177e4
LT		 1203}
1204
463c84b9 1205static inline int inet_csk_reqsk_queue_len(const struct sock *sk)
1da177e4 1206{
463c84b9 1207 return reqsk_queue_len(&inet_csk(sk)->icsk_accept_queue);
1da177e4
LT		 1208}
1209
463c84b9 1210static inline int inet_csk_reqsk_queue_young(const struct sock *sk)
1da177e4 1211{
463c84b9 1212 return reqsk_queue_len_young(&inet_csk(sk)->icsk_accept_queue);
1da177e4
LT		 1213}
1214
463c84b9 1215static inline int inet_csk_reqsk_queue_is_full(const struct sock *sk)
1da177e4 1216{
463c84b9 1217 return reqsk_queue_is_full(&inet_csk(sk)->icsk_accept_queue);
1da177e4
LT		 1218}
1219
463c84b9
ACM		 1220static inline void inet_csk_reqsk_queue_unlink(struct sock *sk,
1221 struct request_sock *req,
1222 struct request_sock **prev)
1da177e4 1223{
463c84b9 1224 reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req, prev);
1da177e4
LT		 1225}
1226
463c84b9
ACM		 1227static inline void inet_csk_reqsk_queue_drop(struct sock *sk,
1228 struct request_sock *req,
1229 struct request_sock **prev)
1da177e4 1230{
463c84b9
ACM		 1231 inet_csk_reqsk_queue_unlink(sk, req, prev);
1232 inet_csk_reqsk_queue_removed(sk, req);
60236fdd 1233 reqsk_free(req);
1da177e4
LT		 1234}
1235
60236fdd 1236static __inline__ void tcp_openreq_init(struct request_sock *req,
1da177e4
LT		 1237 struct tcp_options_received *rx_opt,
1238 struct sk_buff *skb)
1239{
2e6599cb
ACM		 1240 struct inet_request_sock *ireq = inet_rsk(req);
1241
1da177e4 1242 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 1243 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 1244 req->mss = rx_opt->mss_clamp;
1245 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 1246 ireq->tstamp_ok = rx_opt->tstamp_ok;
1247 ireq->sack_ok = rx_opt->sack_ok;
1248 ireq->snd_wscale = rx_opt->snd_wscale;
1249 ireq->wscale_ok = rx_opt->wscale_ok;
1250 ireq->acked = 0;
1251 ireq->ecn_ok = 0;
1252 ireq->rmt_port = skb->h.th->source;
1da177e4
LT		 1253}
1254
1255extern void tcp_enter_memory_pressure(void);
1256
1da177e4
LT		 1257static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1258{
1259 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1260}
1261
1262static inline int keepalive_time_when(const struct tcp_sock *tp)
1263{
1264 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1265}
1266
463c84b9 1267static inline int tcp_fin_time(const struct sock *sk)
1da177e4 1268{
463c84b9
ACM		 1269 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1270 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 1271
463c84b9
ACM		 1272 if (fin_timeout < (rto << 2) - (rto >> 1))
1273 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 1274
1275 return fin_timeout;
1276}
1277
1278static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1279{
1280 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1281 return 0;
1282 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1283 return 0;
1284
1285 /* RST segments are not recommended to carry timestamp,
1286 and, if they do, it is recommended to ignore PAWS because
1287 "their cleanup function should take precedence over timestamps."
1288 Certainly, it is mistake. It is necessary to understand the reasons
1289 of this constraint to relax it: if peer reboots, clock may go
1290 out-of-sync and half-open connections will not be reset.
1291 Actually, the problem would be not existing if all
1292 the implementations followed draft about maintaining clock
1293 via reboots. Linux-2.2 DOES NOT!
1294
1295 However, we can relax time bounds for RST segments to MSL.
1296 */
1297 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1298 return 0;
1299 return 1;
1300}
1301
1da177e4
LT		 1302#define TCP_CHECK_TIMER(sk) do { } while (0)
1303
1304static inline int tcp_use_frto(const struct sock *sk)
1305{
1306 const struct tcp_sock *tp = tcp_sk(sk);
1307
1308 /* F-RTO must be activated in sysctl and there must be some
1309 * unsent new data, and the advertised window should allow
1310 * sending it.
1311 */
1312 return (sysctl_tcp_frto && sk->sk_send_head &&
1313 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1314 tp->snd_una + tp->snd_wnd));
1315}
1316
1317static inline void tcp_mib_init(void)
1318{
1319 /* See RFC 2012 */
1320 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1321 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1322 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1323 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1324}
1325
1326/* /proc */
1327enum tcp_seq_states {
1328 TCP_SEQ_STATE_LISTENING,
1329 TCP_SEQ_STATE_OPENREQ,
1330 TCP_SEQ_STATE_ESTABLISHED,
1331 TCP_SEQ_STATE_TIME_WAIT,
1332};
1333
1334struct tcp_seq_afinfo {
1335 struct module *owner;
1336 char *name;
1337 sa_family_t family;
1338 int (*seq_show) (struct seq_file *m, void *v);
1339 struct file_operations *seq_fops;
1340};
1341
1342struct tcp_iter_state {
1343 sa_family_t family;
1344 enum tcp_seq_states state;
1345 struct sock *syn_wait_sk;
1346 int bucket, sbucket, num, uid;
1347 struct seq_operations seq_ops;
1348};
1349
1350extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1351extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1352
1da177e4 1353#endif /* _TCP_H */
kernel source for telekom puls (alcatel/mediatek)