projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[TCP]: MTU probing
[deliverable/linux.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
22#define FASTRETRANS_DEBUG 1
23
1da177e4
LT		 24#include <linux/config.h>
25#include <linux/list.h>
26#include <linux/tcp.h>
27#include <linux/slab.h>
28#include <linux/cache.h>
29#include <linux/percpu.h>
fb286bb2 30#include <linux/skbuff.h>
3f421baa
ACM		 31
32#include <net/inet_connection_sock.h>
295ff7ed 33#include <net/inet_timewait_sock.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 46extern atomic_t tcp_orphan_count;
1da177e4 47extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 48
1da177e4
LT		 49#define MAX_TCP_HEADER (128 + MAX_HEADER)
50
51/*
52 * Never offer a window over 32767 without using window scaling. Some
53 * poor stacks do signed 16bit maths!
54 */
55#define MAX_TCP_WINDOW 32767U
56
57/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
58#define TCP_MIN_MSS 88U
59
60/* Minimal RCV_MSS. */
61#define TCP_MIN_RCVMSS 536U
62
5d424d5a
JH		 63/* The least MTU to use for probing */
64#define TCP_BASE_MSS 512
65
1da177e4
LT		 66/* After receiving this amount of duplicate ACKs fast retransmit starts. */
67#define TCP_FASTRETRANS_THRESH 3
68
69/* Maximal reordering. */
70#define TCP_MAX_REORDERING 127
71
72/* Maximal number of ACKs sent quickly to accelerate slow-start. */
73#define TCP_MAX_QUICKACKS 16U
74
75/* urg_data states */
76#define TCP_URG_VALID 0x0100
77#define TCP_URG_NOTYET 0x0200
78#define TCP_URG_READ 0x0400
79
80#define TCP_RETR1 3 /*
81 * This is how many retries it does before it
82 * tries to figure out if the gateway is
83 * down. Minimal RFC value is 3; it corresponds
84 * to ~3sec-8min depending on RTO.
85 */
86
87#define TCP_RETR2 15 /*
88 * This should take at least
89 * 90 minutes to time out.
90 * RFC1122 says that the limit is 100 sec.
91 * 15 is ~13-30min depending on RTO.
92 */
93
94#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
caa20d9a 95 * connection: ~180sec is RFC minimum */
1da177e4
LT		 96
97#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
caa20d9a 98 * connection: ~180sec is RFC minimum */
1da177e4
LT		 99
100
101#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
102 * socket. 7 is ~50sec-16min.
103 */
104
105
106#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
112 * TIME-WAIT timer.
113 */
114
115#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
116#if HZ >= 100
117#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118#define TCP_ATO_MIN ((unsigned)(HZ/25))
119#else
120#define TCP_DELACK_MIN 4U
121#define TCP_ATO_MIN 4U
122#endif
123#define TCP_RTO_MAX ((unsigned)(120*HZ))
124#define TCP_RTO_MIN ((unsigned)(HZ/5))
125#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
126
127#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
128 * for local resources.
129 */
130
131#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
132#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
133#define TCP_KEEPALIVE_INTVL (75*HZ)
134
135#define MAX_TCP_KEEPIDLE 32767
136#define MAX_TCP_KEEPINTVL 32767
137#define MAX_TCP_KEEPCNT 127
138#define MAX_TCP_SYNCNT 127
139
140#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
141#define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
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 */
1da177e4
LT		 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
166/*
167 * TCP option lengths
168 */
169
170#define TCPOLEN_MSS 4
171#define TCPOLEN_WINDOW 3
172#define TCPOLEN_SACK_PERM 2
173#define TCPOLEN_TIMESTAMP 10
174
175/* But this is what stacks really send out. */
176#define TCPOLEN_TSTAMP_ALIGNED 12
177#define TCPOLEN_WSCALE_ALIGNED 4
178#define TCPOLEN_SACKPERM_ALIGNED 4
179#define TCPOLEN_SACK_BASE 2
180#define TCPOLEN_SACK_BASE_ALIGNED 4
181#define TCPOLEN_SACK_PERBLOCK 8
182
1da177e4
LT		 183/* Flags in tp->nonagle */
184#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
185#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 186#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 187
295ff7ed
ACM		 188extern struct inet_timewait_death_row tcp_death_row;
189
1da177e4 190/* sysctl variables for tcp */
1da177e4
LT		 191extern int sysctl_tcp_timestamps;
192extern int sysctl_tcp_window_scaling;
193extern int sysctl_tcp_sack;
194extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 195extern int sysctl_tcp_keepalive_time;
196extern int sysctl_tcp_keepalive_probes;
197extern int sysctl_tcp_keepalive_intvl;
198extern int sysctl_tcp_syn_retries;
199extern int sysctl_tcp_synack_retries;
200extern int sysctl_tcp_retries1;
201extern int sysctl_tcp_retries2;
202extern int sysctl_tcp_orphan_retries;
203extern int sysctl_tcp_syncookies;
204extern int sysctl_tcp_retrans_collapse;
205extern int sysctl_tcp_stdurg;
206extern int sysctl_tcp_rfc1337;
207extern int sysctl_tcp_abort_on_overflow;
208extern int sysctl_tcp_max_orphans;
1da177e4
LT		 209extern int sysctl_tcp_fack;
210extern int sysctl_tcp_reordering;
211extern int sysctl_tcp_ecn;
212extern int sysctl_tcp_dsack;
213extern int sysctl_tcp_mem[3];
214extern int sysctl_tcp_wmem[3];
215extern int sysctl_tcp_rmem[3];
216extern int sysctl_tcp_app_win;
217extern int sysctl_tcp_adv_win_scale;
218extern int sysctl_tcp_tw_reuse;
219extern int sysctl_tcp_frto;
220extern int sysctl_tcp_low_latency;
1da177e4 221extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 222extern int sysctl_tcp_moderate_rcvbuf;
223extern int sysctl_tcp_tso_win_divisor;
9772efb9 224extern int sysctl_tcp_abc;
5d424d5a
JH		 225extern int sysctl_tcp_mtu_probing;
226extern int sysctl_tcp_base_mss;
1da177e4
LT		 227
228extern atomic_t tcp_memory_allocated;
229extern atomic_t tcp_sockets_allocated;
230extern int tcp_memory_pressure;
231
1da177e4
LT		 232/*
233 * The next routines deal with comparing 32 bit unsigned ints
234 * and worry about wraparound (automatic with unsigned arithmetic).
235 */
236
237static inline int before(__u32 seq1, __u32 seq2)
238{
239 return (__s32)(seq1-seq2) < 0;
240}
241
242static inline int after(__u32 seq1, __u32 seq2)
243{
244 return (__s32)(seq2-seq1) < 0;
245}
246
247
248/* is s2<=s1<=s3 ? */
249static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
250{
251 return seq3 - seq2 >= seq1 - seq2;
252}
253
254
255extern struct proto tcp_prot;
256
257DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
258#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
259#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
260#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
261#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
262#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
263#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
264
1da177e4
LT		 265extern void tcp_v4_err(struct sk_buff *skb, u32);
266
267extern void tcp_shutdown (struct sock *sk, int how);
268
269extern int tcp_v4_rcv(struct sk_buff *skb);
270
271extern int tcp_v4_remember_stamp(struct sock *sk);
272
8feaf0c0 273extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4
LT		 274
275extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
276 struct msghdr *msg, size_t size);
277extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
278
279extern int tcp_ioctl(struct sock *sk,
280 int cmd,
281 unsigned long arg);
282
283extern int tcp_rcv_state_process(struct sock *sk,
284 struct sk_buff *skb,
285 struct tcphdr *th,
286 unsigned len);
287
288extern int tcp_rcv_established(struct sock *sk,
289 struct sk_buff *skb,
290 struct tcphdr *th,
291 unsigned len);
292
293extern void tcp_rcv_space_adjust(struct sock *sk);
294
6d6ee43e
ACM		 295extern int tcp_twsk_unique(struct sock *sk,
296 struct sock *sktw, void *twp);
297
463c84b9
ACM		 298static inline void tcp_dec_quickack_mode(struct sock *sk,
299 const unsigned int pkts)
1da177e4 300{
463c84b9 301 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 302
463c84b9
ACM		 303 if (icsk->icsk_ack.quick) {
304 if (pkts >= icsk->icsk_ack.quick) {
305 icsk->icsk_ack.quick = 0;
fc6415bc 306 /* Leaving quickack mode we deflate ATO. */
463c84b9 307 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 308 } else
463c84b9 309 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 310 }
311}
312
463c84b9 313extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 314
1da177e4
LT		 315static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
316{
317 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
318}
319
320enum tcp_tw_status
321{
322 TCP_TW_SUCCESS = 0,
323 TCP_TW_RST = 1,
324 TCP_TW_ACK = 2,
325 TCP_TW_SYN = 3
326};
327
328
8feaf0c0 329extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 330 struct sk_buff *skb,
8feaf0c0 331 const struct tcphdr *th);
1da177e4
LT		 332
333extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 334 struct request_sock *req,
335 struct request_sock **prev);
1da177e4
LT		 336extern int tcp_child_process(struct sock *parent,
337 struct sock *child,
338 struct sk_buff *skb);
339extern void tcp_enter_frto(struct sock *sk);
340extern void tcp_enter_loss(struct sock *sk, int how);
341extern void tcp_clear_retrans(struct tcp_sock *tp);
342extern void tcp_update_metrics(struct sock *sk);
343
344extern void tcp_close(struct sock *sk,
345 long timeout);
1da177e4
LT		 346extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
347
348extern int tcp_getsockopt(struct sock *sk, int level,
349 int optname,
350 char __user *optval,
351 int __user *optlen);
352extern int tcp_setsockopt(struct sock *sk, int level,
353 int optname, char __user *optval,
354 int optlen);
355extern void tcp_set_keepalive(struct sock *sk, int val);
356extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
357 struct msghdr *msg,
358 size_t len, int nonblock,
359 int flags, int *addr_len);
360
1da177e4
LT		 361extern void tcp_parse_options(struct sk_buff *skb,
362 struct tcp_options_received *opt_rx,
363 int estab);
364
365/*
366 * TCP v4 functions exported for the inet6 API
367 */
368
8292a17a 369extern void tcp_v4_send_check(struct sock *sk, int len,
1da177e4
LT		 370 struct sk_buff *skb);
371
372extern int tcp_v4_conn_request(struct sock *sk,
373 struct sk_buff *skb);
374
375extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 376 struct request_sock *req,
1da177e4
LT		 377 struct sk_buff *skb);
378
379extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
380 struct sk_buff *skb,
60236fdd 381 struct request_sock *req,
1da177e4
LT		 382 struct dst_entry *dst);
383
384extern int tcp_v4_do_rcv(struct sock *sk,
385 struct sk_buff *skb);
386
387extern int tcp_v4_connect(struct sock *sk,
388 struct sockaddr *uaddr,
389 int addr_len);
390
391extern int tcp_connect(struct sock *sk);
392
393extern struct sk_buff * tcp_make_synack(struct sock *sk,
394 struct dst_entry *dst,
60236fdd 395 struct request_sock *req);
1da177e4
LT		 396
397extern int tcp_disconnect(struct sock *sk, int flags);
398
399extern void tcp_unhash(struct sock *sk);
400
401extern int tcp_v4_hash_connecting(struct sock *sk);
402
403
404/* From syncookies.c */
405extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
406 struct ip_options *opt);
407extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
408 __u16 *mss);
409
410/* tcp_output.c */
411
f6302d1d 412extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
a2e2a59c 413 unsigned int cur_mss, int nonagle);
f6302d1d 414extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
1da177e4
LT		 415extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
416extern void tcp_xmit_retransmit_queue(struct sock *);
417extern void tcp_simple_retransmit(struct sock *);
418extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 419extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 420
421extern void tcp_send_probe0(struct sock *);
422extern void tcp_send_partial(struct sock *);
423extern int tcp_write_wakeup(struct sock *);
424extern void tcp_send_fin(struct sock *sk);
dd0fc66f 425extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
1da177e4 426extern int tcp_send_synack(struct sock *);
c1b4a7e6 427extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 428extern void tcp_send_ack(struct sock *sk);
429extern void tcp_send_delayed_ack(struct sock *sk);
430
a762a980
DM		 431/* tcp_input.c */
432extern void tcp_cwnd_application_limited(struct sock *sk);
433
1da177e4
LT		 434/* tcp_timer.c */
435extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 436static inline void tcp_clear_xmit_timers(struct sock *sk)
437{
438 inet_csk_clear_xmit_timers(sk);
439}
1da177e4 440
1da177e4
LT		 441extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
442extern unsigned int tcp_current_mss(struct sock *sk, int large);
443
17b085ea 444/* tcp.c */
1da177e4
LT		 445extern void tcp_get_info(struct sock *, struct tcp_info *);
446
447/* Read 'sendfile()'-style from a TCP socket */
448typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
449 unsigned int, size_t);
450extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
451 sk_read_actor_t recv_actor);
452
40efc6fa 453extern void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 454
5d424d5a
JH		 455extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
456extern int tcp_mss_to_mtu(struct sock *sk, int mss);
457extern void tcp_mtup_init(struct sock *sk);
458
40efc6fa 459static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 460{
461 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
462 ntohl(TCP_FLAG_ACK) |
463 snd_wnd);
464}
465
40efc6fa 466static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 467{
468 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
469}
470
471static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
472{
b03efcfb 473 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 474 tp->rcv_wnd &&
475 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
476 !tp->urg_data)
477 tcp_fast_path_on(tp);
478}
479
480/* Compute the actual receive window we are currently advertising.
481 * Rcv_nxt can be after the window if our peer push more data
482 * than the offered window.
483 */
40efc6fa 484static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 485{
486 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
487
488 if (win < 0)
489 win = 0;
490 return (u32) win;
491}
492
493/* Choose a new window, without checks for shrinking, and without
494 * scaling applied to the result. The caller does these things
495 * if necessary. This is a "raw" window selection.
496 */
497extern u32 __tcp_select_window(struct sock *sk);
498
499/* TCP timestamps are only 32-bits, this causes a slight
500 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 501 * of jiffies in the buffer control blocks below. We decided
502 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 503 * casts with the following macro.
504 */
505#define tcp_time_stamp ((__u32)(jiffies))
506
caa20d9a 507/* This is what the send packet queuing engine uses to pass
1da177e4
LT		 508 * TCP per-packet control information to the transmission
509 * code. We also store the host-order sequence numbers in
510 * here too. This is 36 bytes on 32-bit architectures,
511 * 40 bytes on 64-bit machines, if this grows please adjust
512 * skbuff.h:skbuff->cb[xxx] size appropriately.
513 */
514struct tcp_skb_cb {
515 union {
516 struct inet_skb_parm h4;
517#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
518 struct inet6_skb_parm h6;
519#endif
520 } header; /* For incoming frames */
521 __u32 seq; /* Starting sequence number */
522 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
523 __u32 when; /* used to compute rtt's */
524 __u8 flags; /* TCP header flags. */
525
526 /* NOTE: These must match up to the flags byte in a
527 * real TCP header.
528 */
529#define TCPCB_FLAG_FIN 0x01
530#define TCPCB_FLAG_SYN 0x02
531#define TCPCB_FLAG_RST 0x04
532#define TCPCB_FLAG_PSH 0x08
533#define TCPCB_FLAG_ACK 0x10
534#define TCPCB_FLAG_URG 0x20
535#define TCPCB_FLAG_ECE 0x40
536#define TCPCB_FLAG_CWR 0x80
537
538 __u8 sacked; /* State flags for SACK/FACK. */
539#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
540#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
541#define TCPCB_LOST 0x04 /* SKB is lost */
542#define TCPCB_TAGBITS 0x07 /* All tag bits */
543
544#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
545#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
546
caa20d9a 547#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
1da177e4
LT		 548
549#define TCPCB_AT_TAIL (TCPCB_URG)
550
551 __u16 urg_ptr; /* Valid w/URG flags is set. */
552 __u32 ack_seq; /* Sequence number ACK'd */
553};
554
555#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
556
557#include <net/tcp_ecn.h>
558
559/* Due to TSO, an SKB can be composed of multiple actual
560 * packets. To keep these tracked properly, we use this.
561 */
562static inline int tcp_skb_pcount(const struct sk_buff *skb)
563{
564 return skb_shinfo(skb)->tso_segs;
565}
566
567/* This is valid iff tcp_skb_pcount() > 1. */
568static inline int tcp_skb_mss(const struct sk_buff *skb)
569{
570 return skb_shinfo(skb)->tso_size;
571}
572
573static inline void tcp_dec_pcount_approx(__u32 *count,
574 const struct sk_buff *skb)
575{
576 if (*count) {
577 *count -= tcp_skb_pcount(skb);
578 if ((int)*count < 0)
579 *count = 0;
580 }
581}
582
583static inline void tcp_packets_out_inc(struct sock *sk,
584 struct tcp_sock *tp,
585 const struct sk_buff *skb)
586{
587 int orig = tp->packets_out;
588
589 tp->packets_out += tcp_skb_pcount(skb);
590 if (!orig)
3f421baa
ACM		 591 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
592 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 593}
594
595static inline void tcp_packets_out_dec(struct tcp_sock *tp,
596 const struct sk_buff *skb)
597{
598 tp->packets_out -= tcp_skb_pcount(skb);
599}
600
317a76f9
SH		 601/* Events passed to congestion control interface */
602enum tcp_ca_event {
603 CA_EVENT_TX_START, /* first transmit when no packets in flight */
604 CA_EVENT_CWND_RESTART, /* congestion window restart */
605 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
606 CA_EVENT_FRTO, /* fast recovery timeout */
607 CA_EVENT_LOSS, /* loss timeout */
608 CA_EVENT_FAST_ACK, /* in sequence ack */
609 CA_EVENT_SLOW_ACK, /* other ack */
610};
611
612/*
613 * Interface for adding new TCP congestion control handlers
614 */
615#define TCP_CA_NAME_MAX 16
616struct tcp_congestion_ops {
617 struct list_head list;
618
619 /* initialize private data (optional) */
6687e988 620 void (*init)(struct sock *sk);
317a76f9 621 /* cleanup private data (optional) */
6687e988 622 void (*release)(struct sock *sk);
317a76f9
SH		 623
624 /* return slow start threshold (required) */
6687e988 625 u32 (*ssthresh)(struct sock *sk);
317a76f9 626 /* lower bound for congestion window (optional) */
6687e988 627 u32 (*min_cwnd)(struct sock *sk);
317a76f9 628 /* do new cwnd calculation (required) */
6687e988 629 void (*cong_avoid)(struct sock *sk, u32 ack,
317a76f9
SH		 630 u32 rtt, u32 in_flight, int good_ack);
631 /* round trip time sample per acked packet (optional) */
6687e988 632 void (*rtt_sample)(struct sock *sk, u32 usrtt);
317a76f9 633 /* call before changing ca_state (optional) */
6687e988 634 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 635 /* call when cwnd event occurs (optional) */
6687e988 636 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 637 /* new value of cwnd after loss (optional) */
6687e988 638 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 639 /* hook for packet ack accounting (optional) */
6687e988 640 void (*pkts_acked)(struct sock *sk, u32 num_acked);
73c1f4a0 641 /* get info for inet_diag (optional) */
6687e988 642 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 643
644 char name[TCP_CA_NAME_MAX];
645 struct module *owner;
646};
647
648extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
649extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
650
6687e988
ACM		 651extern void tcp_init_congestion_control(struct sock *sk);
652extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 653extern int tcp_set_default_congestion_control(const char *name);
654extern void tcp_get_default_congestion_control(char *name);
6687e988 655extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 656extern void tcp_slow_start(struct tcp_sock *tp);
317a76f9 657
5f8ef48d 658extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988
ACM		 659extern u32 tcp_reno_ssthresh(struct sock *sk);
660extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
317a76f9 661 u32 rtt, u32 in_flight, int flag);
6687e988 662extern u32 tcp_reno_min_cwnd(struct sock *sk);
a8acfbac 663extern struct tcp_congestion_ops tcp_reno;
317a76f9 664
6687e988 665static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 666{
6687e988
ACM		 667 struct inet_connection_sock *icsk = inet_csk(sk);
668
669 if (icsk->icsk_ca_ops->set_state)
670 icsk->icsk_ca_ops->set_state(sk, ca_state);
671 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 672}
673
6687e988 674static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 675{
6687e988
ACM		 676 const struct inet_connection_sock *icsk = inet_csk(sk);
677
678 if (icsk->icsk_ca_ops->cwnd_event)
679 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 680}
681
1da177e4
LT		 682/* This determines how many packets are "in the network" to the best
683 * of our knowledge. In many cases it is conservative, but where
684 * detailed information is available from the receiver (via SACK
685 * blocks etc.) we can make more aggressive calculations.
686 *
687 * Use this for decisions involving congestion control, use just
688 * tp->packets_out to determine if the send queue is empty or not.
689 *
690 * Read this equation as:
691 *
692 * "Packets sent once on transmission queue" MINUS
693 * "Packets left network, but not honestly ACKed yet" PLUS
694 * "Packets fast retransmitted"
695 */
40efc6fa 696static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4
LT		 697{
698 return (tp->packets_out - tp->left_out + tp->retrans_out);
699}
700
1da177e4
LT		 701/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
702 * The exception is rate halving phase, when cwnd is decreasing towards
703 * ssthresh.
704 */
6687e988 705static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 706{
6687e988
ACM		 707 const struct tcp_sock *tp = tcp_sk(sk);
708 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 709 return tp->snd_ssthresh;
710 else
711 return max(tp->snd_ssthresh,
712 ((tp->snd_cwnd >> 1) +
713 (tp->snd_cwnd >> 2)));
714}
715
716static inline void tcp_sync_left_out(struct tcp_sock *tp)
717{
718 if (tp->rx_opt.sack_ok &&
719 (tp->sacked_out >= tp->packets_out - tp->lost_out))
720 tp->sacked_out = tp->packets_out - tp->lost_out;
721 tp->left_out = tp->sacked_out + tp->lost_out;
722}
723
40efc6fa 724extern void tcp_enter_cwr(struct sock *sk);
1da177e4
LT		 725extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
726
727/* Slow start with delack produces 3 packets of burst, so that
728 * it is safe "de facto".
729 */
730static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
731{
732 return 3;
733}
734
f4805ede
SH		 735/* RFC2861 Check whether we are limited by application or congestion window
736 * This is the inverse of cwnd check in tcp_tso_should_defer
737 */
738static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
739{
740 const struct tcp_sock *tp = tcp_sk(sk);
741 u32 left;
742
743 if (in_flight >= tp->snd_cwnd)
744 return 1;
745
746 if (!(sk->sk_route_caps & NETIF_F_TSO))
747 return 0;
748
749 left = tp->snd_cwnd - in_flight;
750 if (sysctl_tcp_tso_win_divisor)
751 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
752 else
753 return left <= tcp_max_burst(tp);
754}
755
40efc6fa
SH		 756static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
757 const struct sk_buff *skb)
1da177e4
LT		 758{
759 if (skb->len < mss)
760 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
761}
762
40efc6fa 763static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
1da177e4 764{
463c84b9
ACM		 765 const struct inet_connection_sock *icsk = inet_csk(sk);
766 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 767 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
768 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 769}
770
40efc6fa
SH		 771static inline void tcp_push_pending_frames(struct sock *sk,
772 struct tcp_sock *tp)
1da177e4
LT		 773{
774 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
775}
776
40efc6fa 777static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 778{
779 tp->snd_wl1 = seq;
780}
781
40efc6fa 782static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 783{
784 tp->snd_wl1 = seq;
785}
786
1da177e4
LT		 787/*
788 * Calculate(/check) TCP checksum
789 */
40efc6fa
SH		 790static inline u16 tcp_v4_check(struct tcphdr *th, int len,
791 unsigned long saddr, unsigned long daddr,
792 unsigned long base)
1da177e4
LT		 793{
794 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
795}
796
40efc6fa 797static inline int __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 798{
fb286bb2 799 return __skb_checksum_complete(skb);
1da177e4
LT		 800}
801
40efc6fa 802static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4
LT		 803{
804 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
805 __tcp_checksum_complete(skb);
806}
807
808/* Prequeue for VJ style copy to user, combined with checksumming. */
809
40efc6fa 810static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 811{
812 tp->ucopy.task = NULL;
813 tp->ucopy.len = 0;
814 tp->ucopy.memory = 0;
815 skb_queue_head_init(&tp->ucopy.prequeue);
816}
817
818/* Packet is added to VJ-style prequeue for processing in process
819 * context, if a reader task is waiting. Apparently, this exciting
820 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
821 * failed somewhere. Latency? Burstiness? Well, at least now we will
822 * see, why it failed. 8)8) --ANK
823 *
824 * NOTE: is this not too big to inline?
825 */
40efc6fa 826static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 827{
828 struct tcp_sock *tp = tcp_sk(sk);
829
830 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
831 __skb_queue_tail(&tp->ucopy.prequeue, skb);
832 tp->ucopy.memory += skb->truesize;
833 if (tp->ucopy.memory > sk->sk_rcvbuf) {
834 struct sk_buff *skb1;
835
836 BUG_ON(sock_owned_by_user(sk));
837
838 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
839 sk->sk_backlog_rcv(sk, skb1);
840 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
841 }
842
843 tp->ucopy.memory = 0;
844 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
845 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 846 if (!inet_csk_ack_scheduled(sk))
847 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
3f421baa
ACM		 848 (3 * TCP_RTO_MIN) / 4,
849 TCP_RTO_MAX);
1da177e4
LT		 850 }
851 return 1;
852 }
853 return 0;
854}
855
856
857#undef STATE_TRACE
858
859#ifdef STATE_TRACE
860static const char *statename[]={
861 "Unused","Established","Syn Sent","Syn Recv",
862 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
863 "Close Wait","Last ACK","Listen","Closing"
864};
865#endif
866
40efc6fa 867static inline void tcp_set_state(struct sock *sk, int state)
1da177e4
LT		 868{
869 int oldstate = sk->sk_state;
870
871 switch (state) {
872 case TCP_ESTABLISHED:
873 if (oldstate != TCP_ESTABLISHED)
874 TCP_INC_STATS(TCP_MIB_CURRESTAB);
875 break;
876
877 case TCP_CLOSE:
878 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
879 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
880
881 sk->sk_prot->unhash(sk);
463c84b9 882 if (inet_csk(sk)->icsk_bind_hash &&
1da177e4 883 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2d8c4ce5 884 inet_put_port(&tcp_hashinfo, sk);
1da177e4
LT		 885 /* fall through */
886 default:
887 if (oldstate==TCP_ESTABLISHED)
888 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
889 }
890
891 /* Change state AFTER socket is unhashed to avoid closed
892 * socket sitting in hash tables.
893 */
894 sk->sk_state = state;
895
896#ifdef STATE_TRACE
897 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
898#endif
899}
900
40efc6fa 901static inline void tcp_done(struct sock *sk)
1da177e4
LT		 902{
903 tcp_set_state(sk, TCP_CLOSE);
904 tcp_clear_xmit_timers(sk);
905
906 sk->sk_shutdown = SHUTDOWN_MASK;
907
908 if (!sock_flag(sk, SOCK_DEAD))
909 sk->sk_state_change(sk);
910 else
0a5578cf 911 inet_csk_destroy_sock(sk);
1da177e4
LT		 912}
913
40efc6fa 914static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 915{
916 rx_opt->dsack = 0;
917 rx_opt->eff_sacks = 0;
918 rx_opt->num_sacks = 0;
919}
920
1da177e4
LT		 921/* Determine a window scaling and initial window to offer. */
922extern void tcp_select_initial_window(int __space, __u32 mss,
923 __u32 *rcv_wnd, __u32 *window_clamp,
924 int wscale_ok, __u8 *rcv_wscale);
925
926static inline int tcp_win_from_space(int space)
927{
928 return sysctl_tcp_adv_win_scale<=0 ?
929 (space>>(-sysctl_tcp_adv_win_scale)) :
930 space - (space>>sysctl_tcp_adv_win_scale);
931}
932
933/* Note: caller must be prepared to deal with negative returns */
934static inline int tcp_space(const struct sock *sk)
935{
936 return tcp_win_from_space(sk->sk_rcvbuf -
937 atomic_read(&sk->sk_rmem_alloc));
938}
939
940static inline int tcp_full_space(const struct sock *sk)
941{
942 return tcp_win_from_space(sk->sk_rcvbuf);
943}
944
40efc6fa
SH		 945static inline void tcp_openreq_init(struct request_sock *req,
946 struct tcp_options_received *rx_opt,
947 struct sk_buff *skb)
1da177e4 948{
2e6599cb
ACM		 949 struct inet_request_sock *ireq = inet_rsk(req);
950
1da177e4 951 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 952 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 953 req->mss = rx_opt->mss_clamp;
954 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 955 ireq->tstamp_ok = rx_opt->tstamp_ok;
956 ireq->sack_ok = rx_opt->sack_ok;
957 ireq->snd_wscale = rx_opt->snd_wscale;
958 ireq->wscale_ok = rx_opt->wscale_ok;
959 ireq->acked = 0;
960 ireq->ecn_ok = 0;
961 ireq->rmt_port = skb->h.th->source;
1da177e4
LT		 962}
963
964extern void tcp_enter_memory_pressure(void);
965
1da177e4
LT		 966static inline int keepalive_intvl_when(const struct tcp_sock *tp)
967{
968 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
969}
970
971static inline int keepalive_time_when(const struct tcp_sock *tp)
972{
973 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
974}
975
463c84b9 976static inline int tcp_fin_time(const struct sock *sk)
1da177e4 977{
463c84b9
ACM		 978 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
979 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 980
463c84b9
ACM		 981 if (fin_timeout < (rto << 2) - (rto >> 1))
982 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 983
984 return fin_timeout;
985}
986
987static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
988{
989 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
990 return 0;
991 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
992 return 0;
993
994 /* RST segments are not recommended to carry timestamp,
995 and, if they do, it is recommended to ignore PAWS because
996 "their cleanup function should take precedence over timestamps."
997 Certainly, it is mistake. It is necessary to understand the reasons
998 of this constraint to relax it: if peer reboots, clock may go
999 out-of-sync and half-open connections will not be reset.
1000 Actually, the problem would be not existing if all
1001 the implementations followed draft about maintaining clock
1002 via reboots. Linux-2.2 DOES NOT!
1003
1004 However, we can relax time bounds for RST segments to MSL.
1005 */
1006 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1007 return 0;
1008 return 1;
1009}
1010
1da177e4
LT		 1011#define TCP_CHECK_TIMER(sk) do { } while (0)
1012
1013static inline int tcp_use_frto(const struct sock *sk)
1014{
1015 const struct tcp_sock *tp = tcp_sk(sk);
1016
1017 /* F-RTO must be activated in sysctl and there must be some
1018 * unsent new data, and the advertised window should allow
1019 * sending it.
1020 */
1021 return (sysctl_tcp_frto && sk->sk_send_head &&
1022 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1023 tp->snd_una + tp->snd_wnd));
1024}
1025
1026static inline void tcp_mib_init(void)
1027{
1028 /* See RFC 2012 */
1029 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1030 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1031 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1032 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1033}
1034
6a438bbe
SH		 1035/*from STCP */
1036static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1037 tp->lost_skb_hint = NULL;
1038 tp->scoreboard_skb_hint = NULL;
1039 tp->retransmit_skb_hint = NULL;
1040 tp->forward_skb_hint = NULL;
1041 tp->fastpath_skb_hint = NULL;
1042}
1043
1da177e4
LT		 1044/* /proc */
1045enum tcp_seq_states {
1046 TCP_SEQ_STATE_LISTENING,
1047 TCP_SEQ_STATE_OPENREQ,
1048 TCP_SEQ_STATE_ESTABLISHED,
1049 TCP_SEQ_STATE_TIME_WAIT,
1050};
1051
1052struct tcp_seq_afinfo {
1053 struct module *owner;
1054 char *name;
1055 sa_family_t family;
1056 int (*seq_show) (struct seq_file *m, void *v);
1057 struct file_operations *seq_fops;
1058};
1059
1060struct tcp_iter_state {
1061 sa_family_t family;
1062 enum tcp_seq_states state;
1063 struct sock *syn_wait_sk;
1064 int bucket, sbucket, num, uid;
1065 struct seq_operations seq_ops;
1066};
1067
1068extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1069extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1070
20380731
ACM		 1071extern struct request_sock_ops tcp_request_sock_ops;
1072
1073extern int tcp_v4_destroy_sock(struct sock *sk);
1074
1075#ifdef CONFIG_PROC_FS
1076extern int tcp4_proc_init(void);
1077extern void tcp4_proc_exit(void);
1078#endif
1079
1080extern void tcp_v4_init(struct net_proto_family *ops);
1081extern void tcp_init(void);
1082
1da177e4 1083#endif /* _TCP_H */
Linux kernel - Deliverables
This page took 0.22432 seconds and 5 git commands to generate.