blob: 0f984801197229c2a3280f7a165cfec233233a63 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#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>
Herbert Xufb286bb2005-11-10 13:01:24 -080030#include <linux/skbuff.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070031
32#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070033#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070034#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070036#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <net/sock.h>
38#include <net/snmp.h>
39#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070040#include <net/tcp_states.h>
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/seq_file.h>
43
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070044extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046extern atomic_t tcp_orphan_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047extern void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#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
63/* After receiving this amount of duplicate ACKs fast retransmit starts. */
64#define TCP_FASTRETRANS_THRESH 3
65
66/* Maximal reordering. */
67#define TCP_MAX_REORDERING 127
68
69/* Maximal number of ACKs sent quickly to accelerate slow-start. */
70#define TCP_MAX_QUICKACKS 16U
71
72/* urg_data states */
73#define TCP_URG_VALID 0x0100
74#define TCP_URG_NOTYET 0x0200
75#define TCP_URG_READ 0x0400
76
77#define TCP_RETR1 3 /*
78 * This is how many retries it does before it
79 * tries to figure out if the gateway is
80 * down. Minimal RFC value is 3; it corresponds
81 * to ~3sec-8min depending on RTO.
82 */
83
84#define TCP_RETR2 15 /*
85 * This should take at least
86 * 90 minutes to time out.
87 * RFC1122 says that the limit is 100 sec.
88 * 15 is ~13-30min depending on RTO.
89 */
90
91#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080092 * connection: ~180sec is RFC minimum */
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
94#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080095 * connection: ~180sec is RFC minimum */
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97
98#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
99 * socket. 7 is ~50sec-16min.
100 */
101
102
103#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
104 * state, about 60 seconds */
105#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
106 /* BSD style FIN_WAIT2 deadlock breaker.
107 * It used to be 3min, new value is 60sec,
108 * to combine FIN-WAIT-2 timeout with
109 * TIME-WAIT timer.
110 */
111
112#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
113#if HZ >= 100
114#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
115#define TCP_ATO_MIN ((unsigned)(HZ/25))
116#else
117#define TCP_DELACK_MIN 4U
118#define TCP_ATO_MIN 4U
119#endif
120#define TCP_RTO_MAX ((unsigned)(120*HZ))
121#define TCP_RTO_MIN ((unsigned)(HZ/5))
122#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
123
124#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
125 * for local resources.
126 */
127
128#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
129#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
130#define TCP_KEEPALIVE_INTVL (75*HZ)
131
132#define MAX_TCP_KEEPIDLE 32767
133#define MAX_TCP_KEEPINTVL 32767
134#define MAX_TCP_KEEPCNT 127
135#define MAX_TCP_SYNCNT 127
136
137#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
138#define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
139
140#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
141#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
142 * after this time. It should be equal
143 * (or greater than) TCP_TIMEWAIT_LEN
144 * to provide reliability equal to one
145 * provided by timewait state.
146 */
147#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
148 * timestamps. It must be less than
149 * minimal timewait lifetime.
150 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151/*
152 * TCP option
153 */
154
155#define TCPOPT_NOP 1 /* Padding */
156#define TCPOPT_EOL 0 /* End of options */
157#define TCPOPT_MSS 2 /* Segment size negotiating */
158#define TCPOPT_WINDOW 3 /* Window scaling */
159#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
160#define TCPOPT_SACK 5 /* SACK Block */
161#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
162
163/*
164 * TCP option lengths
165 */
166
167#define TCPOLEN_MSS 4
168#define TCPOLEN_WINDOW 3
169#define TCPOLEN_SACK_PERM 2
170#define TCPOLEN_TIMESTAMP 10
171
172/* But this is what stacks really send out. */
173#define TCPOLEN_TSTAMP_ALIGNED 12
174#define TCPOLEN_WSCALE_ALIGNED 4
175#define TCPOLEN_SACKPERM_ALIGNED 4
176#define TCPOLEN_SACK_BASE 2
177#define TCPOLEN_SACK_BASE_ALIGNED 4
178#define TCPOLEN_SACK_PERBLOCK 8
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180/* Flags in tp->nonagle */
181#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
182#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800183#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700185extern struct inet_timewait_death_row tcp_death_row;
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188extern int sysctl_tcp_timestamps;
189extern int sysctl_tcp_window_scaling;
190extern int sysctl_tcp_sack;
191extern int sysctl_tcp_fin_timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192extern int sysctl_tcp_keepalive_time;
193extern int sysctl_tcp_keepalive_probes;
194extern int sysctl_tcp_keepalive_intvl;
195extern int sysctl_tcp_syn_retries;
196extern int sysctl_tcp_synack_retries;
197extern int sysctl_tcp_retries1;
198extern int sysctl_tcp_retries2;
199extern int sysctl_tcp_orphan_retries;
200extern int sysctl_tcp_syncookies;
201extern int sysctl_tcp_retrans_collapse;
202extern int sysctl_tcp_stdurg;
203extern int sysctl_tcp_rfc1337;
204extern int sysctl_tcp_abort_on_overflow;
205extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206extern int sysctl_tcp_fack;
207extern int sysctl_tcp_reordering;
208extern int sysctl_tcp_ecn;
209extern int sysctl_tcp_dsack;
210extern int sysctl_tcp_mem[3];
211extern int sysctl_tcp_wmem[3];
212extern int sysctl_tcp_rmem[3];
213extern int sysctl_tcp_app_win;
214extern int sysctl_tcp_adv_win_scale;
215extern int sysctl_tcp_tw_reuse;
216extern int sysctl_tcp_frto;
217extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219extern int sysctl_tcp_moderate_rcvbuf;
220extern int sysctl_tcp_tso_win_divisor;
Stephen Hemminger9772efb2005-11-10 17:09:53 -0800221extern int sysctl_tcp_abc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223extern atomic_t tcp_memory_allocated;
224extern atomic_t tcp_sockets_allocated;
225extern int tcp_memory_pressure;
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227/*
228 * Pointers to address related TCP functions
229 * (i.e. things that depend on the address family)
230 */
231
232struct tcp_func {
233 int (*queue_xmit) (struct sk_buff *skb,
234 int ipfragok);
235
236 void (*send_check) (struct sock *sk,
237 struct tcphdr *th,
238 int len,
239 struct sk_buff *skb);
240
241 int (*rebuild_header) (struct sock *sk);
242
243 int (*conn_request) (struct sock *sk,
244 struct sk_buff *skb);
245
246 struct sock * (*syn_recv_sock) (struct sock *sk,
247 struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700248 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 struct dst_entry *dst);
250
251 int (*remember_stamp) (struct sock *sk);
252
253 __u16 net_header_len;
254
255 int (*setsockopt) (struct sock *sk,
256 int level,
257 int optname,
258 char __user *optval,
259 int optlen);
260
261 int (*getsockopt) (struct sock *sk,
262 int level,
263 int optname,
264 char __user *optval,
265 int __user *optlen);
266
267
268 void (*addr2sockaddr) (struct sock *sk,
269 struct sockaddr *);
270
271 int sockaddr_len;
272};
273
274/*
275 * The next routines deal with comparing 32 bit unsigned ints
276 * and worry about wraparound (automatic with unsigned arithmetic).
277 */
278
279static inline int before(__u32 seq1, __u32 seq2)
280{
281 return (__s32)(seq1-seq2) < 0;
282}
283
284static inline int after(__u32 seq1, __u32 seq2)
285{
286 return (__s32)(seq2-seq1) < 0;
287}
288
289
290/* is s2<=s1<=s3 ? */
291static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
292{
293 return seq3 - seq2 >= seq1 - seq2;
294}
295
296
297extern struct proto tcp_prot;
298
299DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
300#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
301#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
302#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
303#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
304#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
305#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307extern void tcp_v4_err(struct sk_buff *skb, u32);
308
309extern void tcp_shutdown (struct sock *sk, int how);
310
311extern int tcp_v4_rcv(struct sk_buff *skb);
312
313extern int tcp_v4_remember_stamp(struct sock *sk);
314
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700315extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
318 struct msghdr *msg, size_t size);
319extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
320
321extern int tcp_ioctl(struct sock *sk,
322 int cmd,
323 unsigned long arg);
324
325extern int tcp_rcv_state_process(struct sock *sk,
326 struct sk_buff *skb,
327 struct tcphdr *th,
328 unsigned len);
329
330extern int tcp_rcv_established(struct sock *sk,
331 struct sk_buff *skb,
332 struct tcphdr *th,
333 unsigned len);
334
335extern void tcp_rcv_space_adjust(struct sock *sk);
336
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700337static inline void tcp_dec_quickack_mode(struct sock *sk,
338 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700340 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700341
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700342 if (icsk->icsk_ack.quick) {
343 if (pkts >= icsk->icsk_ack.quick) {
344 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700345 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700346 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700347 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700348 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 }
350}
351
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700352extern void tcp_enter_quickack_mode(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
355{
356 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
357}
358
359enum tcp_tw_status
360{
361 TCP_TW_SUCCESS = 0,
362 TCP_TW_RST = 1,
363 TCP_TW_ACK = 2,
364 TCP_TW_SYN = 3
365};
366
367
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700368extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 struct sk_buff *skb,
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700370 const struct tcphdr *th);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700373 struct request_sock *req,
374 struct request_sock **prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375extern int tcp_child_process(struct sock *parent,
376 struct sock *child,
377 struct sk_buff *skb);
378extern void tcp_enter_frto(struct sock *sk);
379extern void tcp_enter_loss(struct sock *sk, int how);
380extern void tcp_clear_retrans(struct tcp_sock *tp);
381extern void tcp_update_metrics(struct sock *sk);
382
383extern void tcp_close(struct sock *sk,
384 long timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
386
387extern int tcp_getsockopt(struct sock *sk, int level,
388 int optname,
389 char __user *optval,
390 int __user *optlen);
391extern int tcp_setsockopt(struct sock *sk, int level,
392 int optname, char __user *optval,
393 int optlen);
394extern void tcp_set_keepalive(struct sock *sk, int val);
395extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
396 struct msghdr *msg,
397 size_t len, int nonblock,
398 int flags, int *addr_len);
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400extern void tcp_parse_options(struct sk_buff *skb,
401 struct tcp_options_received *opt_rx,
402 int estab);
403
404/*
405 * TCP v4 functions exported for the inet6 API
406 */
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408extern void tcp_v4_send_check(struct sock *sk,
409 struct tcphdr *th, int len,
410 struct sk_buff *skb);
411
412extern int tcp_v4_conn_request(struct sock *sk,
413 struct sk_buff *skb);
414
415extern struct sock * tcp_create_openreq_child(struct sock *sk,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700416 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 struct sk_buff *skb);
418
419extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
420 struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700421 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 struct dst_entry *dst);
423
424extern int tcp_v4_do_rcv(struct sock *sk,
425 struct sk_buff *skb);
426
427extern int tcp_v4_connect(struct sock *sk,
428 struct sockaddr *uaddr,
429 int addr_len);
430
431extern int tcp_connect(struct sock *sk);
432
433extern struct sk_buff * tcp_make_synack(struct sock *sk,
434 struct dst_entry *dst,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700435 struct request_sock *req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437extern int tcp_disconnect(struct sock *sk, int flags);
438
439extern void tcp_unhash(struct sock *sk);
440
441extern int tcp_v4_hash_connecting(struct sock *sk);
442
443
444/* From syncookies.c */
445extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
446 struct ip_options *opt);
447extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
448 __u16 *mss);
449
450/* tcp_output.c */
451
David S. Millerf6302d12005-07-05 15:18:03 -0700452extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
David S. Millera2e2a592005-07-05 15:19:23 -0700453 unsigned int cur_mss, int nonagle);
David S. Millerf6302d12005-07-05 15:18:03 -0700454extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
456extern void tcp_xmit_retransmit_queue(struct sock *);
457extern void tcp_simple_retransmit(struct sock *);
458extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
David S. Miller6475be12005-09-01 22:47:01 -0700459extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
461extern void tcp_send_probe0(struct sock *);
462extern void tcp_send_partial(struct sock *);
463extern int tcp_write_wakeup(struct sock *);
464extern void tcp_send_fin(struct sock *sk);
Al Virodd0fc662005-10-07 07:46:04 +0100465extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466extern int tcp_send_synack(struct sock *);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700467extern void tcp_push_one(struct sock *, unsigned int mss_now);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468extern void tcp_send_ack(struct sock *sk);
469extern void tcp_send_delayed_ack(struct sock *sk);
470
David S. Millera762a982005-07-05 15:18:51 -0700471/* tcp_input.c */
472extern void tcp_cwnd_application_limited(struct sock *sk);
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/* tcp_timer.c */
475extern void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700476static inline void tcp_clear_xmit_timers(struct sock *sk)
477{
478 inet_csk_clear_xmit_timers(sk);
479}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
482extern unsigned int tcp_current_mss(struct sock *sk, int large);
483
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300484/* tcp.c */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485extern void tcp_get_info(struct sock *, struct tcp_info *);
486
487/* Read 'sendfile()'-style from a TCP socket */
488typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
489 unsigned int, size_t);
490extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
491 sk_read_actor_t recv_actor);
492
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493/* Initialize RCV_MSS value.
494 * RCV_MSS is an our guess about MSS used by the peer.
495 * We haven't any direct information about the MSS.
496 * It's better to underestimate the RCV_MSS rather than overestimate.
497 * Overestimations make us ACKing less frequently than needed.
498 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
499 */
500
501static inline void tcp_initialize_rcv_mss(struct sock *sk)
502{
503 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700504 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506 hint = min(hint, tp->rcv_wnd/2);
507 hint = min(hint, TCP_MIN_RCVMSS);
508 hint = max(hint, TCP_MIN_MSS);
509
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700510 inet_csk(sk)->icsk_ack.rcv_mss = hint;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
513static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
514{
515 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
516 ntohl(TCP_FLAG_ACK) |
517 snd_wnd);
518}
519
520static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
521{
522 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
523}
524
525static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
526{
David S. Millerb03efcf2005-07-08 14:57:23 -0700527 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 tp->rcv_wnd &&
529 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
530 !tp->urg_data)
531 tcp_fast_path_on(tp);
532}
533
534/* Compute the actual receive window we are currently advertising.
535 * Rcv_nxt can be after the window if our peer push more data
536 * than the offered window.
537 */
538static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
539{
540 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
541
542 if (win < 0)
543 win = 0;
544 return (u32) win;
545}
546
547/* Choose a new window, without checks for shrinking, and without
548 * scaling applied to the result. The caller does these things
549 * if necessary. This is a "raw" window selection.
550 */
551extern u32 __tcp_select_window(struct sock *sk);
552
553/* TCP timestamps are only 32-bits, this causes a slight
554 * complication on 64-bit systems since we store a snapshot
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800555 * of jiffies in the buffer control blocks below. We decidedly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 * only use of the low 32-bits of jiffies and hide the ugly
557 * casts with the following macro.
558 */
559#define tcp_time_stamp ((__u32)(jiffies))
560
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800561/* This is what the send packet queuing engine uses to pass
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * TCP per-packet control information to the transmission
563 * code. We also store the host-order sequence numbers in
564 * here too. This is 36 bytes on 32-bit architectures,
565 * 40 bytes on 64-bit machines, if this grows please adjust
566 * skbuff.h:skbuff->cb[xxx] size appropriately.
567 */
568struct tcp_skb_cb {
569 union {
570 struct inet_skb_parm h4;
571#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
572 struct inet6_skb_parm h6;
573#endif
574 } header; /* For incoming frames */
575 __u32 seq; /* Starting sequence number */
576 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
577 __u32 when; /* used to compute rtt's */
578 __u8 flags; /* TCP header flags. */
579
580 /* NOTE: These must match up to the flags byte in a
581 * real TCP header.
582 */
583#define TCPCB_FLAG_FIN 0x01
584#define TCPCB_FLAG_SYN 0x02
585#define TCPCB_FLAG_RST 0x04
586#define TCPCB_FLAG_PSH 0x08
587#define TCPCB_FLAG_ACK 0x10
588#define TCPCB_FLAG_URG 0x20
589#define TCPCB_FLAG_ECE 0x40
590#define TCPCB_FLAG_CWR 0x80
591
592 __u8 sacked; /* State flags for SACK/FACK. */
593#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
594#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
595#define TCPCB_LOST 0x04 /* SKB is lost */
596#define TCPCB_TAGBITS 0x07 /* All tag bits */
597
598#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
599#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
600
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800601#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603#define TCPCB_AT_TAIL (TCPCB_URG)
604
605 __u16 urg_ptr; /* Valid w/URG flags is set. */
606 __u32 ack_seq; /* Sequence number ACK'd */
607};
608
609#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
610
611#include <net/tcp_ecn.h>
612
613/* Due to TSO, an SKB can be composed of multiple actual
614 * packets. To keep these tracked properly, we use this.
615 */
616static inline int tcp_skb_pcount(const struct sk_buff *skb)
617{
618 return skb_shinfo(skb)->tso_segs;
619}
620
621/* This is valid iff tcp_skb_pcount() > 1. */
622static inline int tcp_skb_mss(const struct sk_buff *skb)
623{
624 return skb_shinfo(skb)->tso_size;
625}
626
627static inline void tcp_dec_pcount_approx(__u32 *count,
628 const struct sk_buff *skb)
629{
630 if (*count) {
631 *count -= tcp_skb_pcount(skb);
632 if ((int)*count < 0)
633 *count = 0;
634 }
635}
636
637static inline void tcp_packets_out_inc(struct sock *sk,
638 struct tcp_sock *tp,
639 const struct sk_buff *skb)
640{
641 int orig = tp->packets_out;
642
643 tp->packets_out += tcp_skb_pcount(skb);
644 if (!orig)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700645 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
646 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
649static inline void tcp_packets_out_dec(struct tcp_sock *tp,
650 const struct sk_buff *skb)
651{
652 tp->packets_out -= tcp_skb_pcount(skb);
653}
654
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700655/* Events passed to congestion control interface */
656enum tcp_ca_event {
657 CA_EVENT_TX_START, /* first transmit when no packets in flight */
658 CA_EVENT_CWND_RESTART, /* congestion window restart */
659 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
660 CA_EVENT_FRTO, /* fast recovery timeout */
661 CA_EVENT_LOSS, /* loss timeout */
662 CA_EVENT_FAST_ACK, /* in sequence ack */
663 CA_EVENT_SLOW_ACK, /* other ack */
664};
665
666/*
667 * Interface for adding new TCP congestion control handlers
668 */
669#define TCP_CA_NAME_MAX 16
670struct tcp_congestion_ops {
671 struct list_head list;
672
673 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300674 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700675 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300676 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700677
678 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300679 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700680 /* lower bound for congestion window (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300681 u32 (*min_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700682 /* do new cwnd calculation (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300683 void (*cong_avoid)(struct sock *sk, u32 ack,
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700684 u32 rtt, u32 in_flight, int good_ack);
685 /* round trip time sample per acked packet (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300686 void (*rtt_sample)(struct sock *sk, u32 usrtt);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700687 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300688 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700689 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300690 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700691 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300692 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700693 /* hook for packet ack accounting (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300694 void (*pkts_acked)(struct sock *sk, u32 num_acked);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300695 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300696 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700697
698 char name[TCP_CA_NAME_MAX];
699 struct module *owner;
700};
701
702extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
703extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
704
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300705extern void tcp_init_congestion_control(struct sock *sk);
706extern void tcp_cleanup_congestion_control(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700707extern int tcp_set_default_congestion_control(const char *name);
708extern void tcp_get_default_congestion_control(char *name);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300709extern int tcp_set_congestion_control(struct sock *sk, const char *name);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700710
Stephen Hemminger5f8ef482005-06-23 20:37:36 -0700711extern struct tcp_congestion_ops tcp_init_congestion_ops;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300712extern u32 tcp_reno_ssthresh(struct sock *sk);
713extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700714 u32 rtt, u32 in_flight, int flag);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300715extern u32 tcp_reno_min_cwnd(struct sock *sk);
David S. Millera8acfba2005-06-23 23:45:02 -0700716extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700717
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300718static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700719{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300720 struct inet_connection_sock *icsk = inet_csk(sk);
721
722 if (icsk->icsk_ca_ops->set_state)
723 icsk->icsk_ca_ops->set_state(sk, ca_state);
724 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700725}
726
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300727static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700728{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300729 const struct inet_connection_sock *icsk = inet_csk(sk);
730
731 if (icsk->icsk_ca_ops->cwnd_event)
732 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735/* This determines how many packets are "in the network" to the best
736 * of our knowledge. In many cases it is conservative, but where
737 * detailed information is available from the receiver (via SACK
738 * blocks etc.) we can make more aggressive calculations.
739 *
740 * Use this for decisions involving congestion control, use just
741 * tp->packets_out to determine if the send queue is empty or not.
742 *
743 * Read this equation as:
744 *
745 * "Packets sent once on transmission queue" MINUS
746 * "Packets left network, but not honestly ACKed yet" PLUS
747 * "Packets fast retransmitted"
748 */
749static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
750{
751 return (tp->packets_out - tp->left_out + tp->retrans_out);
752}
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
755 * The exception is rate halving phase, when cwnd is decreasing towards
756 * ssthresh.
757 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300758static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300760 const struct tcp_sock *tp = tcp_sk(sk);
761 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return tp->snd_ssthresh;
763 else
764 return max(tp->snd_ssthresh,
765 ((tp->snd_cwnd >> 1) +
766 (tp->snd_cwnd >> 2)));
767}
768
Stephen Hemminger7faffa12005-11-10 17:07:24 -0800769/*
770 * Linear increase during slow start
771 */
772static inline void tcp_slow_start(struct tcp_sock *tp)
773{
Stephen Hemminger9772efb2005-11-10 17:09:53 -0800774 if (sysctl_tcp_abc) {
775 /* RFC3465: Slow Start
776 * TCP sender SHOULD increase cwnd by the number of
777 * previously unacknowledged bytes ACKed by each incoming
778 * acknowledgment, provided the increase is not more than L
779 */
780 if (tp->bytes_acked < tp->mss_cache)
781 return;
782
783 /* We MAY increase by 2 if discovered delayed ack */
784 if (sysctl_tcp_abc > 1 && tp->bytes_acked > 2*tp->mss_cache) {
785 if (tp->snd_cwnd < tp->snd_cwnd_clamp)
786 tp->snd_cwnd++;
787 }
788 }
789 tp->bytes_acked = 0;
790
Stephen Hemminger7faffa12005-11-10 17:07:24 -0800791 if (tp->snd_cwnd < tp->snd_cwnd_clamp)
792 tp->snd_cwnd++;
793}
794
795
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796static inline void tcp_sync_left_out(struct tcp_sock *tp)
797{
798 if (tp->rx_opt.sack_ok &&
799 (tp->sacked_out >= tp->packets_out - tp->lost_out))
800 tp->sacked_out = tp->packets_out - tp->lost_out;
801 tp->left_out = tp->sacked_out + tp->lost_out;
802}
803
Baruch Evend1b04c02005-07-30 17:41:59 -0700804/* Set slow start threshold and cwnd not falling to slow start */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300805static inline void __tcp_enter_cwr(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300807 const struct inet_connection_sock *icsk = inet_csk(sk);
808 struct tcp_sock *tp = tcp_sk(sk);
809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 tp->undo_marker = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300811 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 tp->snd_cwnd = min(tp->snd_cwnd,
813 tcp_packets_in_flight(tp) + 1U);
814 tp->snd_cwnd_cnt = 0;
815 tp->high_seq = tp->snd_nxt;
816 tp->snd_cwnd_stamp = tcp_time_stamp;
817 TCP_ECN_queue_cwr(tp);
818}
819
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300820static inline void tcp_enter_cwr(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300822 struct tcp_sock *tp = tcp_sk(sk);
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 tp->prior_ssthresh = 0;
Stephen Hemminger9772efb2005-11-10 17:09:53 -0800825 tp->bytes_acked = 0;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300826 if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {
827 __tcp_enter_cwr(sk);
828 tcp_set_ca_state(sk, TCP_CA_CWR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
830}
831
832extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
833
834/* Slow start with delack produces 3 packets of burst, so that
835 * it is safe "de facto".
836 */
837static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
838{
839 return 3;
840}
841
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800842/* RFC2861 Check whether we are limited by application or congestion window
843 * This is the inverse of cwnd check in tcp_tso_should_defer
844 */
845static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
846{
847 const struct tcp_sock *tp = tcp_sk(sk);
848 u32 left;
849
850 if (in_flight >= tp->snd_cwnd)
851 return 1;
852
853 if (!(sk->sk_route_caps & NETIF_F_TSO))
854 return 0;
855
856 left = tp->snd_cwnd - in_flight;
857 if (sysctl_tcp_tso_win_divisor)
858 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
859 else
860 return left <= tcp_max_burst(tp);
861}
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
864 const struct sk_buff *skb)
865{
866 if (skb->len < mss)
867 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
868}
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
871{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700872 const struct inet_connection_sock *icsk = inet_csk(sk);
873 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700874 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
875 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876}
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878static __inline__ void tcp_push_pending_frames(struct sock *sk,
879 struct tcp_sock *tp)
880{
881 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
882}
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
885{
886 tp->snd_wl1 = seq;
887}
888
889static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
890{
891 tp->snd_wl1 = seq;
892}
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894/*
895 * Calculate(/check) TCP checksum
896 */
897static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
898 unsigned long saddr, unsigned long daddr,
899 unsigned long base)
900{
901 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
902}
903
904static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
905{
Herbert Xufb286bb2005-11-10 13:01:24 -0800906 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
909static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
910{
911 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
912 __tcp_checksum_complete(skb);
913}
914
915/* Prequeue for VJ style copy to user, combined with checksumming. */
916
917static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
918{
919 tp->ucopy.task = NULL;
920 tp->ucopy.len = 0;
921 tp->ucopy.memory = 0;
922 skb_queue_head_init(&tp->ucopy.prequeue);
923}
924
925/* Packet is added to VJ-style prequeue for processing in process
926 * context, if a reader task is waiting. Apparently, this exciting
927 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
928 * failed somewhere. Latency? Burstiness? Well, at least now we will
929 * see, why it failed. 8)8) --ANK
930 *
931 * NOTE: is this not too big to inline?
932 */
933static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
934{
935 struct tcp_sock *tp = tcp_sk(sk);
936
937 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
938 __skb_queue_tail(&tp->ucopy.prequeue, skb);
939 tp->ucopy.memory += skb->truesize;
940 if (tp->ucopy.memory > sk->sk_rcvbuf) {
941 struct sk_buff *skb1;
942
943 BUG_ON(sock_owned_by_user(sk));
944
945 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
946 sk->sk_backlog_rcv(sk, skb1);
947 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
948 }
949
950 tp->ucopy.memory = 0;
951 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
952 wake_up_interruptible(sk->sk_sleep);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700953 if (!inet_csk_ack_scheduled(sk))
954 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700955 (3 * TCP_RTO_MIN) / 4,
956 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
958 return 1;
959 }
960 return 0;
961}
962
963
964#undef STATE_TRACE
965
966#ifdef STATE_TRACE
967static const char *statename[]={
968 "Unused","Established","Syn Sent","Syn Recv",
969 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
970 "Close Wait","Last ACK","Listen","Closing"
971};
972#endif
973
974static __inline__ void tcp_set_state(struct sock *sk, int state)
975{
976 int oldstate = sk->sk_state;
977
978 switch (state) {
979 case TCP_ESTABLISHED:
980 if (oldstate != TCP_ESTABLISHED)
981 TCP_INC_STATS(TCP_MIB_CURRESTAB);
982 break;
983
984 case TCP_CLOSE:
985 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
986 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
987
988 sk->sk_prot->unhash(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700989 if (inet_csk(sk)->icsk_bind_hash &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700991 inet_put_port(&tcp_hashinfo, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 /* fall through */
993 default:
994 if (oldstate==TCP_ESTABLISHED)
995 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
996 }
997
998 /* Change state AFTER socket is unhashed to avoid closed
999 * socket sitting in hash tables.
1000 */
1001 sk->sk_state = state;
1002
1003#ifdef STATE_TRACE
1004 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1005#endif
1006}
1007
1008static __inline__ void tcp_done(struct sock *sk)
1009{
1010 tcp_set_state(sk, TCP_CLOSE);
1011 tcp_clear_xmit_timers(sk);
1012
1013 sk->sk_shutdown = SHUTDOWN_MASK;
1014
1015 if (!sock_flag(sk, SOCK_DEAD))
1016 sk->sk_state_change(sk);
1017 else
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -07001018 inet_csk_destroy_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019}
1020
1021static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1022{
1023 rx_opt->dsack = 0;
1024 rx_opt->eff_sacks = 0;
1025 rx_opt->num_sacks = 0;
1026}
1027
1028static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1029{
1030 if (tp->rx_opt.tstamp_ok) {
1031 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1032 (TCPOPT_NOP << 16) |
1033 (TCPOPT_TIMESTAMP << 8) |
1034 TCPOLEN_TIMESTAMP);
1035 *ptr++ = htonl(tstamp);
1036 *ptr++ = htonl(tp->rx_opt.ts_recent);
1037 }
1038 if (tp->rx_opt.eff_sacks) {
1039 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1040 int this_sack;
1041
1042 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1043 (TCPOPT_NOP << 16) |
1044 (TCPOPT_SACK << 8) |
1045 (TCPOLEN_SACK_BASE +
1046 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1047 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1048 *ptr++ = htonl(sp[this_sack].start_seq);
1049 *ptr++ = htonl(sp[this_sack].end_seq);
1050 }
1051 if (tp->rx_opt.dsack) {
1052 tp->rx_opt.dsack = 0;
1053 tp->rx_opt.eff_sacks--;
1054 }
1055 }
1056}
1057
1058/* Construct a tcp options header for a SYN or SYN_ACK packet.
1059 * If this is every changed make sure to change the definition of
1060 * MAX_SYN_SIZE to match the new maximum number of options that you
1061 * can generate.
1062 */
1063static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1064 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1065{
1066 /* We always get an MSS option.
1067 * The option bytes which will be seen in normal data
1068 * packets should timestamps be used, must be in the MSS
1069 * advertised. But we subtract them from tp->mss_cache so
1070 * that calculations in tcp_sendmsg are simpler etc.
1071 * So account for this fact here if necessary. If we
1072 * don't do this correctly, as a receiver we won't
1073 * recognize data packets as being full sized when we
1074 * should, and thus we won't abide by the delayed ACK
1075 * rules correctly.
1076 * SACKs don't matter, we never delay an ACK when we
1077 * have any of those going out.
1078 */
1079 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1080 if (ts) {
1081 if(sack)
1082 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1083 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1084 else
1085 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1086 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1087 *ptr++ = htonl(tstamp); /* TSVAL */
1088 *ptr++ = htonl(ts_recent); /* TSECR */
1089 } else if(sack)
1090 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1091 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1092 if (offer_wscale)
1093 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1094}
1095
1096/* Determine a window scaling and initial window to offer. */
1097extern void tcp_select_initial_window(int __space, __u32 mss,
1098 __u32 *rcv_wnd, __u32 *window_clamp,
1099 int wscale_ok, __u8 *rcv_wscale);
1100
1101static inline int tcp_win_from_space(int space)
1102{
1103 return sysctl_tcp_adv_win_scale<=0 ?
1104 (space>>(-sysctl_tcp_adv_win_scale)) :
1105 space - (space>>sysctl_tcp_adv_win_scale);
1106}
1107
1108/* Note: caller must be prepared to deal with negative returns */
1109static inline int tcp_space(const struct sock *sk)
1110{
1111 return tcp_win_from_space(sk->sk_rcvbuf -
1112 atomic_read(&sk->sk_rmem_alloc));
1113}
1114
1115static inline int tcp_full_space(const struct sock *sk)
1116{
1117 return tcp_win_from_space(sk->sk_rcvbuf);
1118}
1119
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -07001120static __inline__ void tcp_openreq_init(struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 struct tcp_options_received *rx_opt,
1122 struct sk_buff *skb)
1123{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001124 struct inet_request_sock *ireq = inet_rsk(req);
1125
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001127 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 req->mss = rx_opt->mss_clamp;
1129 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001130 ireq->tstamp_ok = rx_opt->tstamp_ok;
1131 ireq->sack_ok = rx_opt->sack_ok;
1132 ireq->snd_wscale = rx_opt->snd_wscale;
1133 ireq->wscale_ok = rx_opt->wscale_ok;
1134 ireq->acked = 0;
1135 ireq->ecn_ok = 0;
1136 ireq->rmt_port = skb->h.th->source;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139extern void tcp_enter_memory_pressure(void);
1140
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1142{
1143 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1144}
1145
1146static inline int keepalive_time_when(const struct tcp_sock *tp)
1147{
1148 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1149}
1150
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001151static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001153 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1154 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001156 if (fin_timeout < (rto << 2) - (rto >> 1))
1157 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
1159 return fin_timeout;
1160}
1161
1162static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1163{
1164 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1165 return 0;
1166 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1167 return 0;
1168
1169 /* RST segments are not recommended to carry timestamp,
1170 and, if they do, it is recommended to ignore PAWS because
1171 "their cleanup function should take precedence over timestamps."
1172 Certainly, it is mistake. It is necessary to understand the reasons
1173 of this constraint to relax it: if peer reboots, clock may go
1174 out-of-sync and half-open connections will not be reset.
1175 Actually, the problem would be not existing if all
1176 the implementations followed draft about maintaining clock
1177 via reboots. Linux-2.2 DOES NOT!
1178
1179 However, we can relax time bounds for RST segments to MSL.
1180 */
1181 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1182 return 0;
1183 return 1;
1184}
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186#define TCP_CHECK_TIMER(sk) do { } while (0)
1187
1188static inline int tcp_use_frto(const struct sock *sk)
1189{
1190 const struct tcp_sock *tp = tcp_sk(sk);
1191
1192 /* F-RTO must be activated in sysctl and there must be some
1193 * unsent new data, and the advertised window should allow
1194 * sending it.
1195 */
1196 return (sysctl_tcp_frto && sk->sk_send_head &&
1197 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1198 tp->snd_una + tp->snd_wnd));
1199}
1200
1201static inline void tcp_mib_init(void)
1202{
1203 /* See RFC 2012 */
1204 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1205 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1206 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1207 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1208}
1209
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001210/*from STCP */
1211static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1212 tp->lost_skb_hint = NULL;
1213 tp->scoreboard_skb_hint = NULL;
1214 tp->retransmit_skb_hint = NULL;
1215 tp->forward_skb_hint = NULL;
1216 tp->fastpath_skb_hint = NULL;
1217}
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219/* /proc */
1220enum tcp_seq_states {
1221 TCP_SEQ_STATE_LISTENING,
1222 TCP_SEQ_STATE_OPENREQ,
1223 TCP_SEQ_STATE_ESTABLISHED,
1224 TCP_SEQ_STATE_TIME_WAIT,
1225};
1226
1227struct tcp_seq_afinfo {
1228 struct module *owner;
1229 char *name;
1230 sa_family_t family;
1231 int (*seq_show) (struct seq_file *m, void *v);
1232 struct file_operations *seq_fops;
1233};
1234
1235struct tcp_iter_state {
1236 sa_family_t family;
1237 enum tcp_seq_states state;
1238 struct sock *syn_wait_sk;
1239 int bucket, sbucket, num, uid;
1240 struct seq_operations seq_ops;
1241};
1242
1243extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1244extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1245
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001246extern struct request_sock_ops tcp_request_sock_ops;
1247
1248extern int tcp_v4_destroy_sock(struct sock *sk);
1249
1250#ifdef CONFIG_PROC_FS
1251extern int tcp4_proc_init(void);
1252extern void tcp4_proc_exit(void);
1253#endif
1254
1255extern void tcp_v4_init(struct net_proto_family *ops);
1256extern void tcp_init(void);
1257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258#endif /* _TCP_H */