blob: af9273204cfd446857d5282b52a8ce733f03da79 [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/list.h>
25#include <linux/tcp.h>
26#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
Herbert Xufb286bb2005-11-10 13:01:24 -080029#include <linux/skbuff.h>
Chris Leech97fc2f02006-05-23 17:55:33 -070030#include <linux/dmaengine.h>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080031#include <linux/crypto.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070032
33#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070034#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070035#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070037#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <net/sock.h>
39#include <net/snmp.h>
40#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070041#include <net/tcp_states.h>
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/seq_file.h>
44
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070045extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047extern atomic_t tcp_orphan_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048extern void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#define MAX_TCP_HEADER (128 + MAX_HEADER)
51
52/*
53 * Never offer a window over 32767 without using window scaling. Some
54 * poor stacks do signed 16bit maths!
55 */
56#define MAX_TCP_WINDOW 32767U
57
58/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59#define TCP_MIN_MSS 88U
60
61/* Minimal RCV_MSS. */
62#define TCP_MIN_RCVMSS 536U
63
John Heffner5d424d52006-03-20 17:53:41 -080064/* The least MTU to use for probing */
65#define TCP_BASE_MSS 512
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067/* After receiving this amount of duplicate ACKs fast retransmit starts. */
68#define TCP_FASTRETRANS_THRESH 3
69
70/* Maximal reordering. */
71#define TCP_MAX_REORDERING 127
72
73/* Maximal number of ACKs sent quickly to accelerate slow-start. */
74#define TCP_MAX_QUICKACKS 16U
75
76/* urg_data states */
77#define TCP_URG_VALID 0x0100
78#define TCP_URG_NOTYET 0x0200
79#define TCP_URG_READ 0x0400
80
81#define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
86 */
87
88#define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
93 */
94
95#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080096 * connection: ~180sec is RFC minimum */
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080099 * connection: ~180sec is RFC minimum */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
101
102#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
103 * socket. 7 is ~50sec-16min.
104 */
105
106
107#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
113 * TIME-WAIT timer.
114 */
115
116#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
117#if HZ >= 100
118#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119#define TCP_ATO_MIN ((unsigned)(HZ/25))
120#else
121#define TCP_DELACK_MIN 4U
122#define TCP_ATO_MIN 4U
123#endif
124#define TCP_RTO_MAX ((unsigned)(120*HZ))
125#define TCP_RTO_MIN ((unsigned)(HZ/5))
126#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
127
128#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
129 * for local resources.
130 */
131
132#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
133#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
134#define TCP_KEEPALIVE_INTVL (75*HZ)
135
136#define MAX_TCP_KEEPIDLE 32767
137#define MAX_TCP_KEEPINTVL 32767
138#define MAX_TCP_KEEPCNT 127
139#define MAX_TCP_SYNCNT 127
140
141#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154/*
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 */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800165#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167/*
168 * TCP option lengths
169 */
170
171#define TCPOLEN_MSS 4
172#define TCPOLEN_WINDOW 3
173#define TCPOLEN_SACK_PERM 2
174#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800175#define TCPOLEN_MD5SIG 18
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
177/* But this is what stacks really send out. */
178#define TCPOLEN_TSTAMP_ALIGNED 12
179#define TCPOLEN_WSCALE_ALIGNED 4
180#define TCPOLEN_SACKPERM_ALIGNED 4
181#define TCPOLEN_SACK_BASE 2
182#define TCPOLEN_SACK_BASE_ALIGNED 4
183#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800184#define TCPOLEN_MD5SIG_ALIGNED 20
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/* Flags in tp->nonagle */
187#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
188#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800189#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700191extern struct inet_timewait_death_row tcp_death_row;
192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194extern int sysctl_tcp_timestamps;
195extern int sysctl_tcp_window_scaling;
196extern int sysctl_tcp_sack;
197extern int sysctl_tcp_fin_timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198extern int sysctl_tcp_keepalive_time;
199extern int sysctl_tcp_keepalive_probes;
200extern int sysctl_tcp_keepalive_intvl;
201extern int sysctl_tcp_syn_retries;
202extern int sysctl_tcp_synack_retries;
203extern int sysctl_tcp_retries1;
204extern int sysctl_tcp_retries2;
205extern int sysctl_tcp_orphan_retries;
206extern int sysctl_tcp_syncookies;
207extern int sysctl_tcp_retrans_collapse;
208extern int sysctl_tcp_stdurg;
209extern int sysctl_tcp_rfc1337;
210extern int sysctl_tcp_abort_on_overflow;
211extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212extern int sysctl_tcp_fack;
213extern int sysctl_tcp_reordering;
214extern int sysctl_tcp_ecn;
215extern int sysctl_tcp_dsack;
216extern int sysctl_tcp_mem[3];
217extern int sysctl_tcp_wmem[3];
218extern int sysctl_tcp_rmem[3];
219extern int sysctl_tcp_app_win;
220extern int sysctl_tcp_adv_win_scale;
221extern int sysctl_tcp_tw_reuse;
222extern int sysctl_tcp_frto;
Ilpo Järvinen3cfe3ba2007-02-27 10:09:49 -0800223extern int sysctl_tcp_frto_response;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224extern int sysctl_tcp_low_latency;
Chris Leech95937822006-05-23 18:02:55 -0700225extern int sysctl_tcp_dma_copybreak;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227extern int sysctl_tcp_moderate_rcvbuf;
228extern int sysctl_tcp_tso_win_divisor;
Stephen Hemminger9772efb2005-11-10 17:09:53 -0800229extern int sysctl_tcp_abc;
John Heffner5d424d52006-03-20 17:53:41 -0800230extern int sysctl_tcp_mtu_probing;
231extern int sysctl_tcp_base_mss;
Rick Jones15d99e02006-03-20 22:40:29 -0800232extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700233extern int sysctl_tcp_slow_start_after_idle;
John Heffner886236c2007-03-25 19:21:45 -0700234extern int sysctl_tcp_max_ssthresh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236extern atomic_t tcp_memory_allocated;
237extern atomic_t tcp_sockets_allocated;
238extern int tcp_memory_pressure;
239
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 * The next routines deal with comparing 32 bit unsigned ints
242 * and worry about wraparound (automatic with unsigned arithmetic).
243 */
244
245static inline int before(__u32 seq1, __u32 seq2)
246{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800247 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800249#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
251/* is s2<=s1<=s3 ? */
252static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
253{
254 return seq3 - seq2 >= seq1 - seq2;
255}
256
257
258extern struct proto tcp_prot;
259
260DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
261#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
262#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
263#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
264#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
265#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
266#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268extern void tcp_v4_err(struct sk_buff *skb, u32);
269
270extern void tcp_shutdown (struct sock *sk, int how);
271
272extern int tcp_v4_rcv(struct sk_buff *skb);
273
274extern int tcp_v4_remember_stamp(struct sock *sk);
275
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700276extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
278extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
279 struct msghdr *msg, size_t size);
280extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
281
282extern int tcp_ioctl(struct sock *sk,
283 int cmd,
284 unsigned long arg);
285
286extern int tcp_rcv_state_process(struct sock *sk,
287 struct sk_buff *skb,
288 struct tcphdr *th,
289 unsigned len);
290
291extern int tcp_rcv_established(struct sock *sk,
292 struct sk_buff *skb,
293 struct tcphdr *th,
294 unsigned len);
295
296extern void tcp_rcv_space_adjust(struct sock *sk);
297
Chris Leech0e4b4992006-05-23 18:00:16 -0700298extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
299
Arnaldo Carvalho de Melo6d6ee432005-12-13 23:25:19 -0800300extern int tcp_twsk_unique(struct sock *sk,
301 struct sock *sktw, void *twp);
302
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800303extern void tcp_twsk_destructor(struct sock *sk);
304
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700305static inline void tcp_dec_quickack_mode(struct sock *sk,
306 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700308 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700309
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700310 if (icsk->icsk_ack.quick) {
311 if (pkts >= icsk->icsk_ack.quick) {
312 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700313 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700314 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700315 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700316 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 }
318}
319
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700320extern void tcp_enter_quickack_mode(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
323{
324 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
325}
326
327enum tcp_tw_status
328{
329 TCP_TW_SUCCESS = 0,
330 TCP_TW_RST = 1,
331 TCP_TW_ACK = 2,
332 TCP_TW_SYN = 3
333};
334
335
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700336extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 struct sk_buff *skb,
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700338 const struct tcphdr *th);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700341 struct request_sock *req,
342 struct request_sock **prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343extern int tcp_child_process(struct sock *parent,
344 struct sock *child,
345 struct sk_buff *skb);
Ilpo Järvinen46d0de42007-02-21 23:10:39 -0800346extern int tcp_use_frto(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347extern void tcp_enter_frto(struct sock *sk);
348extern void tcp_enter_loss(struct sock *sk, int how);
349extern void tcp_clear_retrans(struct tcp_sock *tp);
350extern void tcp_update_metrics(struct sock *sk);
351
352extern void tcp_close(struct sock *sk,
353 long timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
355
356extern int tcp_getsockopt(struct sock *sk, int level,
357 int optname,
358 char __user *optval,
359 int __user *optlen);
360extern int tcp_setsockopt(struct sock *sk, int level,
361 int optname, char __user *optval,
362 int optlen);
Dmitry Mishin3fdadf72006-03-20 22:45:21 -0800363extern int compat_tcp_getsockopt(struct sock *sk,
364 int level, int optname,
365 char __user *optval, int __user *optlen);
366extern int compat_tcp_setsockopt(struct sock *sk,
367 int level, int optname,
368 char __user *optval, int optlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369extern void tcp_set_keepalive(struct sock *sk, int val);
370extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
371 struct msghdr *msg,
372 size_t len, int nonblock,
373 int flags, int *addr_len);
374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375extern void tcp_parse_options(struct sk_buff *skb,
376 struct tcp_options_received *opt_rx,
377 int estab);
378
379/*
380 * TCP v4 functions exported for the inet6 API
381 */
382
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800383extern void tcp_v4_send_check(struct sock *sk, int len,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 struct sk_buff *skb);
385
386extern int tcp_v4_conn_request(struct sock *sk,
387 struct sk_buff *skb);
388
389extern struct sock * tcp_create_openreq_child(struct sock *sk,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700390 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 struct sk_buff *skb);
392
393extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
394 struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700395 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 struct dst_entry *dst);
397
398extern int tcp_v4_do_rcv(struct sock *sk,
399 struct sk_buff *skb);
400
401extern int tcp_v4_connect(struct sock *sk,
402 struct sockaddr *uaddr,
403 int addr_len);
404
405extern int tcp_connect(struct sock *sk);
406
407extern struct sk_buff * tcp_make_synack(struct sock *sk,
408 struct dst_entry *dst,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700409 struct request_sock *req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411extern int tcp_disconnect(struct sock *sk, int flags);
412
413extern void tcp_unhash(struct sock *sk);
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415/* From syncookies.c */
416extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
417 struct ip_options *opt);
418extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
419 __u16 *mss);
420
421/* tcp_output.c */
422
David S. Millerf6302d12005-07-05 15:18:03 -0700423extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
David S. Millera2e2a592005-07-05 15:19:23 -0700424 unsigned int cur_mss, int nonagle);
David S. Millerf6302d12005-07-05 15:18:03 -0700425extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
427extern void tcp_xmit_retransmit_queue(struct sock *);
428extern void tcp_simple_retransmit(struct sock *);
429extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
David S. Miller6475be12005-09-01 22:47:01 -0700430extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432extern void tcp_send_probe0(struct sock *);
433extern void tcp_send_partial(struct sock *);
434extern int tcp_write_wakeup(struct sock *);
435extern void tcp_send_fin(struct sock *sk);
Al Virodd0fc662005-10-07 07:46:04 +0100436extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437extern int tcp_send_synack(struct sock *);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700438extern void tcp_push_one(struct sock *, unsigned int mss_now);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439extern void tcp_send_ack(struct sock *sk);
440extern void tcp_send_delayed_ack(struct sock *sk);
441
David S. Millera762a982005-07-05 15:18:51 -0700442/* tcp_input.c */
443extern void tcp_cwnd_application_limited(struct sock *sk);
444
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445/* tcp_timer.c */
446extern void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700447static inline void tcp_clear_xmit_timers(struct sock *sk)
448{
449 inet_csk_clear_xmit_timers(sk);
450}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
453extern unsigned int tcp_current_mss(struct sock *sk, int large);
454
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300455/* tcp.c */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456extern void tcp_get_info(struct sock *, struct tcp_info *);
457
458/* Read 'sendfile()'-style from a TCP socket */
459typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
460 unsigned int, size_t);
461extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
462 sk_read_actor_t recv_actor);
463
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800464extern void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
John Heffner5d424d52006-03-20 17:53:41 -0800466extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
467extern int tcp_mss_to_mtu(struct sock *sk, int mss);
468extern void tcp_mtup_init(struct sock *sk);
469
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800470static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
472 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
473 ntohl(TCP_FLAG_ACK) |
474 snd_wnd);
475}
476
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800477static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
479 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
480}
481
482static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
483{
David S. Millerb03efcf2005-07-08 14:57:23 -0700484 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 tp->rcv_wnd &&
486 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
487 !tp->urg_data)
488 tcp_fast_path_on(tp);
489}
490
491/* Compute the actual receive window we are currently advertising.
492 * Rcv_nxt can be after the window if our peer push more data
493 * than the offered window.
494 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800495static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
497 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
498
499 if (win < 0)
500 win = 0;
501 return (u32) win;
502}
503
504/* Choose a new window, without checks for shrinking, and without
505 * scaling applied to the result. The caller does these things
506 * if necessary. This is a "raw" window selection.
507 */
508extern u32 __tcp_select_window(struct sock *sk);
509
510/* TCP timestamps are only 32-bits, this causes a slight
511 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800512 * of jiffies in the buffer control blocks below. We decided
513 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 * casts with the following macro.
515 */
516#define tcp_time_stamp ((__u32)(jiffies))
517
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800518/* This is what the send packet queuing engine uses to pass
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 * TCP per-packet control information to the transmission
520 * code. We also store the host-order sequence numbers in
521 * here too. This is 36 bytes on 32-bit architectures,
522 * 40 bytes on 64-bit machines, if this grows please adjust
523 * skbuff.h:skbuff->cb[xxx] size appropriately.
524 */
525struct tcp_skb_cb {
526 union {
527 struct inet_skb_parm h4;
528#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
529 struct inet6_skb_parm h6;
530#endif
531 } header; /* For incoming frames */
532 __u32 seq; /* Starting sequence number */
533 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
534 __u32 when; /* used to compute rtt's */
535 __u8 flags; /* TCP header flags. */
536
537 /* NOTE: These must match up to the flags byte in a
538 * real TCP header.
539 */
540#define TCPCB_FLAG_FIN 0x01
541#define TCPCB_FLAG_SYN 0x02
542#define TCPCB_FLAG_RST 0x04
543#define TCPCB_FLAG_PSH 0x08
544#define TCPCB_FLAG_ACK 0x10
545#define TCPCB_FLAG_URG 0x20
546#define TCPCB_FLAG_ECE 0x40
547#define TCPCB_FLAG_CWR 0x80
548
549 __u8 sacked; /* State flags for SACK/FACK. */
550#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
551#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
552#define TCPCB_LOST 0x04 /* SKB is lost */
553#define TCPCB_TAGBITS 0x07 /* All tag bits */
554
555#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
556#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
557
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800558#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560#define TCPCB_AT_TAIL (TCPCB_URG)
561
562 __u16 urg_ptr; /* Valid w/URG flags is set. */
563 __u32 ack_seq; /* Sequence number ACK'd */
564};
565
566#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
567
568#include <net/tcp_ecn.h>
569
570/* Due to TSO, an SKB can be composed of multiple actual
571 * packets. To keep these tracked properly, we use this.
572 */
573static inline int tcp_skb_pcount(const struct sk_buff *skb)
574{
Herbert Xu79671682006-06-22 02:40:14 -0700575 return skb_shinfo(skb)->gso_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576}
577
578/* This is valid iff tcp_skb_pcount() > 1. */
579static inline int tcp_skb_mss(const struct sk_buff *skb)
580{
Herbert Xu79671682006-06-22 02:40:14 -0700581 return skb_shinfo(skb)->gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582}
583
584static inline void tcp_dec_pcount_approx(__u32 *count,
585 const struct sk_buff *skb)
586{
587 if (*count) {
588 *count -= tcp_skb_pcount(skb);
589 if ((int)*count < 0)
590 *count = 0;
591 }
592}
593
594static inline void tcp_packets_out_inc(struct sock *sk,
595 struct tcp_sock *tp,
596 const struct sk_buff *skb)
597{
598 int orig = tp->packets_out;
599
600 tp->packets_out += tcp_skb_pcount(skb);
601 if (!orig)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700602 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
603 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604}
605
606static inline void tcp_packets_out_dec(struct tcp_sock *tp,
607 const struct sk_buff *skb)
608{
609 tp->packets_out -= tcp_skb_pcount(skb);
610}
611
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700612/* Events passed to congestion control interface */
613enum tcp_ca_event {
614 CA_EVENT_TX_START, /* first transmit when no packets in flight */
615 CA_EVENT_CWND_RESTART, /* congestion window restart */
616 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
617 CA_EVENT_FRTO, /* fast recovery timeout */
618 CA_EVENT_LOSS, /* loss timeout */
619 CA_EVENT_FAST_ACK, /* in sequence ack */
620 CA_EVENT_SLOW_ACK, /* other ack */
621};
622
623/*
624 * Interface for adding new TCP congestion control handlers
625 */
626#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800627#define TCP_CA_MAX 128
628#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
629
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700630struct tcp_congestion_ops {
631 struct list_head list;
Stephen Hemmingerce7bc3b2006-11-09 16:35:15 -0800632 int non_restricted;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700633
634 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300635 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700636 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300637 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700638
639 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300640 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700641 /* lower bound for congestion window (optional) */
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700642 u32 (*min_cwnd)(const struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700643 /* do new cwnd calculation (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300644 void (*cong_avoid)(struct sock *sk, u32 ack,
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700645 u32 rtt, u32 in_flight, int good_ack);
646 /* round trip time sample per acked packet (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300647 void (*rtt_sample)(struct sock *sk, u32 usrtt);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700648 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300649 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700650 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300651 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700652 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300653 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700654 /* hook for packet ack accounting (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300655 void (*pkts_acked)(struct sock *sk, u32 num_acked);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300656 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300657 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700658
659 char name[TCP_CA_NAME_MAX];
660 struct module *owner;
661};
662
663extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
664extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
665
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300666extern void tcp_init_congestion_control(struct sock *sk);
667extern void tcp_cleanup_congestion_control(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700668extern int tcp_set_default_congestion_control(const char *name);
669extern void tcp_get_default_congestion_control(char *name);
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800670extern void tcp_get_available_congestion_control(char *buf, size_t len);
Stephen Hemmingerce7bc3b2006-11-09 16:35:15 -0800671extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
672extern int tcp_set_allowed_congestion_control(char *allowed);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300673extern int tcp_set_congestion_control(struct sock *sk, const char *name);
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800674extern void tcp_slow_start(struct tcp_sock *tp);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700675
Stephen Hemminger5f8ef482005-06-23 20:37:36 -0700676extern struct tcp_congestion_ops tcp_init_congestion_ops;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300677extern u32 tcp_reno_ssthresh(struct sock *sk);
678extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700679 u32 rtt, u32 in_flight, int flag);
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700680extern u32 tcp_reno_min_cwnd(const struct sock *sk);
David S. Millera8acfba2005-06-23 23:45:02 -0700681extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700682
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300683static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700684{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300685 struct inet_connection_sock *icsk = inet_csk(sk);
686
687 if (icsk->icsk_ca_ops->set_state)
688 icsk->icsk_ca_ops->set_state(sk, ca_state);
689 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700690}
691
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300692static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700693{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300694 const struct inet_connection_sock *icsk = inet_csk(sk);
695
696 if (icsk->icsk_ca_ops->cwnd_event)
697 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700/* This determines how many packets are "in the network" to the best
701 * of our knowledge. In many cases it is conservative, but where
702 * detailed information is available from the receiver (via SACK
703 * blocks etc.) we can make more aggressive calculations.
704 *
705 * Use this for decisions involving congestion control, use just
706 * tp->packets_out to determine if the send queue is empty or not.
707 *
708 * Read this equation as:
709 *
710 * "Packets sent once on transmission queue" MINUS
711 * "Packets left network, but not honestly ACKed yet" PLUS
712 * "Packets fast retransmitted"
713 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800714static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
716 return (tp->packets_out - tp->left_out + tp->retrans_out);
717}
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
720 * The exception is rate halving phase, when cwnd is decreasing towards
721 * ssthresh.
722 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300723static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300725 const struct tcp_sock *tp = tcp_sk(sk);
726 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return tp->snd_ssthresh;
728 else
729 return max(tp->snd_ssthresh,
730 ((tp->snd_cwnd >> 1) +
731 (tp->snd_cwnd >> 2)));
732}
733
734static inline void tcp_sync_left_out(struct tcp_sock *tp)
735{
736 if (tp->rx_opt.sack_ok &&
737 (tp->sacked_out >= tp->packets_out - tp->lost_out))
738 tp->sacked_out = tp->packets_out - tp->lost_out;
739 tp->left_out = tp->sacked_out + tp->lost_out;
740}
741
Ilpo Järvinen3cfe3ba2007-02-27 10:09:49 -0800742extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
744
745/* Slow start with delack produces 3 packets of burst, so that
746 * it is safe "de facto".
747 */
748static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
749{
750 return 3;
751}
752
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800753/* RFC2861 Check whether we are limited by application or congestion window
754 * This is the inverse of cwnd check in tcp_tso_should_defer
755 */
756static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
757{
758 const struct tcp_sock *tp = tcp_sk(sk);
759 u32 left;
760
761 if (in_flight >= tp->snd_cwnd)
762 return 1;
763
Herbert Xubcd76112006-06-30 13:36:35 -0700764 if (!sk_can_gso(sk))
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800765 return 0;
766
767 left = tp->snd_cwnd - in_flight;
768 if (sysctl_tcp_tso_win_divisor)
769 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
770 else
771 return left <= tcp_max_burst(tp);
772}
773
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800774static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
775 const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776{
777 if (skb->len < mss)
778 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
779}
780
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800781static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700783 const struct inet_connection_sock *icsk = inet_csk(sk);
784 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700785 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
786 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
788
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800789static inline void tcp_push_pending_frames(struct sock *sk,
790 struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791{
792 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
793}
794
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800795static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
797 tp->snd_wl1 = seq;
798}
799
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800800static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
802 tp->snd_wl1 = seq;
803}
804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805/*
806 * Calculate(/check) TCP checksum
807 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800808static inline __sum16 tcp_v4_check(int len, __be32 saddr,
809 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
811 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
812}
813
Al Virob51655b2006-11-14 21:40:42 -0800814static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
Herbert Xufb286bb2005-11-10 13:01:24 -0800816 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800819static inline int tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
821 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
822 __tcp_checksum_complete(skb);
823}
824
825/* Prequeue for VJ style copy to user, combined with checksumming. */
826
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800827static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
829 tp->ucopy.task = NULL;
830 tp->ucopy.len = 0;
831 tp->ucopy.memory = 0;
832 skb_queue_head_init(&tp->ucopy.prequeue);
Chris Leech97fc2f02006-05-23 17:55:33 -0700833#ifdef CONFIG_NET_DMA
834 tp->ucopy.dma_chan = NULL;
835 tp->ucopy.wakeup = 0;
836 tp->ucopy.pinned_list = NULL;
837 tp->ucopy.dma_cookie = 0;
838#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839}
840
841/* Packet is added to VJ-style prequeue for processing in process
842 * context, if a reader task is waiting. Apparently, this exciting
843 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
844 * failed somewhere. Latency? Burstiness? Well, at least now we will
845 * see, why it failed. 8)8) --ANK
846 *
847 * NOTE: is this not too big to inline?
848 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800849static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850{
851 struct tcp_sock *tp = tcp_sk(sk);
852
853 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
854 __skb_queue_tail(&tp->ucopy.prequeue, skb);
855 tp->ucopy.memory += skb->truesize;
856 if (tp->ucopy.memory > sk->sk_rcvbuf) {
857 struct sk_buff *skb1;
858
859 BUG_ON(sock_owned_by_user(sk));
860
861 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
862 sk->sk_backlog_rcv(sk, skb1);
863 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
864 }
865
866 tp->ucopy.memory = 0;
867 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
868 wake_up_interruptible(sk->sk_sleep);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700869 if (!inet_csk_ack_scheduled(sk))
870 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700871 (3 * TCP_RTO_MIN) / 4,
872 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 }
874 return 1;
875 }
876 return 0;
877}
878
879
880#undef STATE_TRACE
881
882#ifdef STATE_TRACE
883static const char *statename[]={
884 "Unused","Established","Syn Sent","Syn Recv",
885 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
886 "Close Wait","Last ACK","Listen","Closing"
887};
888#endif
889
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800890static inline void tcp_set_state(struct sock *sk, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
892 int oldstate = sk->sk_state;
893
894 switch (state) {
895 case TCP_ESTABLISHED:
896 if (oldstate != TCP_ESTABLISHED)
897 TCP_INC_STATS(TCP_MIB_CURRESTAB);
898 break;
899
900 case TCP_CLOSE:
901 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
902 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
903
904 sk->sk_prot->unhash(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700905 if (inet_csk(sk)->icsk_bind_hash &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700907 inet_put_port(&tcp_hashinfo, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 /* fall through */
909 default:
910 if (oldstate==TCP_ESTABLISHED)
911 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
912 }
913
914 /* Change state AFTER socket is unhashed to avoid closed
915 * socket sitting in hash tables.
916 */
917 sk->sk_state = state;
918
919#ifdef STATE_TRACE
920 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
921#endif
922}
923
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800924static inline void tcp_done(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925{
Wei Yongjun3687b1d2006-07-30 20:35:54 -0700926 if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
927 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 tcp_set_state(sk, TCP_CLOSE);
930 tcp_clear_xmit_timers(sk);
931
932 sk->sk_shutdown = SHUTDOWN_MASK;
933
934 if (!sock_flag(sk, SOCK_DEAD))
935 sk->sk_state_change(sk);
936 else
Arnaldo Carvalho de Melo0a5578c2005-08-09 20:11:41 -0700937 inet_csk_destroy_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938}
939
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800940static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
942 rx_opt->dsack = 0;
943 rx_opt->eff_sacks = 0;
944 rx_opt->num_sacks = 0;
945}
946
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947/* Determine a window scaling and initial window to offer. */
948extern void tcp_select_initial_window(int __space, __u32 mss,
949 __u32 *rcv_wnd, __u32 *window_clamp,
950 int wscale_ok, __u8 *rcv_wscale);
951
952static inline int tcp_win_from_space(int space)
953{
954 return sysctl_tcp_adv_win_scale<=0 ?
955 (space>>(-sysctl_tcp_adv_win_scale)) :
956 space - (space>>sysctl_tcp_adv_win_scale);
957}
958
959/* Note: caller must be prepared to deal with negative returns */
960static inline int tcp_space(const struct sock *sk)
961{
962 return tcp_win_from_space(sk->sk_rcvbuf -
963 atomic_read(&sk->sk_rmem_alloc));
964}
965
966static inline int tcp_full_space(const struct sock *sk)
967{
968 return tcp_win_from_space(sk->sk_rcvbuf);
969}
970
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800971static inline void tcp_openreq_init(struct request_sock *req,
972 struct tcp_options_received *rx_opt,
973 struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700975 struct inet_request_sock *ireq = inet_rsk(req);
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700978 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 req->mss = rx_opt->mss_clamp;
980 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700981 ireq->tstamp_ok = rx_opt->tstamp_ok;
982 ireq->sack_ok = rx_opt->sack_ok;
983 ireq->snd_wscale = rx_opt->snd_wscale;
984 ireq->wscale_ok = rx_opt->wscale_ok;
985 ireq->acked = 0;
986 ireq->ecn_ok = 0;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700987 ireq->rmt_port = tcp_hdr(skb)->source;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988}
989
990extern void tcp_enter_memory_pressure(void);
991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992static inline int keepalive_intvl_when(const struct tcp_sock *tp)
993{
994 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
995}
996
997static inline int keepalive_time_when(const struct tcp_sock *tp)
998{
999 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1000}
1001
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001002static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001004 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1005 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001007 if (fin_timeout < (rto << 2) - (rto >> 1))
1008 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 return fin_timeout;
1011}
1012
1013static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1014{
1015 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1016 return 0;
James Morris9d729f72007-03-04 16:12:44 -08001017 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 return 0;
1019
1020 /* RST segments are not recommended to carry timestamp,
1021 and, if they do, it is recommended to ignore PAWS because
1022 "their cleanup function should take precedence over timestamps."
1023 Certainly, it is mistake. It is necessary to understand the reasons
1024 of this constraint to relax it: if peer reboots, clock may go
1025 out-of-sync and half-open connections will not be reset.
1026 Actually, the problem would be not existing if all
1027 the implementations followed draft about maintaining clock
1028 via reboots. Linux-2.2 DOES NOT!
1029
1030 However, we can relax time bounds for RST segments to MSL.
1031 */
James Morris9d729f72007-03-04 16:12:44 -08001032 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return 0;
1034 return 1;
1035}
1036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037#define TCP_CHECK_TIMER(sk) do { } while (0)
1038
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039static inline void tcp_mib_init(void)
1040{
1041 /* See RFC 2012 */
1042 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1043 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1044 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1045 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1046}
1047
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001048/*from STCP */
1049static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1050 tp->lost_skb_hint = NULL;
1051 tp->scoreboard_skb_hint = NULL;
1052 tp->retransmit_skb_hint = NULL;
1053 tp->forward_skb_hint = NULL;
1054 tp->fastpath_skb_hint = NULL;
1055}
1056
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001057/* MD5 Signature */
1058struct crypto_hash;
1059
1060/* - key database */
1061struct tcp_md5sig_key {
1062 u8 *key;
1063 u8 keylen;
1064};
1065
1066struct tcp4_md5sig_key {
1067 u8 *key;
1068 u16 keylen;
1069 __be32 addr;
1070};
1071
1072struct tcp6_md5sig_key {
1073 u8 *key;
1074 u16 keylen;
1075#if 0
1076 u32 scope_id; /* XXX */
1077#endif
1078 struct in6_addr addr;
1079};
1080
1081/* - sock block */
1082struct tcp_md5sig_info {
1083 struct tcp4_md5sig_key *keys4;
1084#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1085 struct tcp6_md5sig_key *keys6;
1086 u32 entries6;
1087 u32 alloced6;
1088#endif
1089 u32 entries4;
1090 u32 alloced4;
1091};
1092
1093/* - pseudo header */
1094struct tcp4_pseudohdr {
1095 __be32 saddr;
1096 __be32 daddr;
1097 __u8 pad;
1098 __u8 protocol;
1099 __be16 len;
1100};
1101
1102struct tcp6_pseudohdr {
1103 struct in6_addr saddr;
1104 struct in6_addr daddr;
1105 __be32 len;
1106 __be32 protocol; /* including padding */
1107};
1108
1109union tcp_md5sum_block {
1110 struct tcp4_pseudohdr ip4;
1111#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1112 struct tcp6_pseudohdr ip6;
1113#endif
1114};
1115
1116/* - pool: digest algorithm, hash description and scratch buffer */
1117struct tcp_md5sig_pool {
1118 struct hash_desc md5_desc;
1119 union tcp_md5sum_block md5_blk;
1120};
1121
1122#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1123
1124/* - functions */
1125extern int tcp_v4_calc_md5_hash(char *md5_hash,
1126 struct tcp_md5sig_key *key,
1127 struct sock *sk,
1128 struct dst_entry *dst,
1129 struct request_sock *req,
1130 struct tcphdr *th,
1131 int protocol, int tcplen);
1132extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1133 struct sock *addr_sk);
1134
1135extern int tcp_v4_md5_do_add(struct sock *sk,
1136 __be32 addr,
1137 u8 *newkey,
1138 u8 newkeylen);
1139
1140extern int tcp_v4_md5_do_del(struct sock *sk,
Al Viro8e5200f2006-11-20 18:06:37 -08001141 __be32 addr);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001142
1143extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1144extern void tcp_free_md5sig_pool(void);
1145
1146extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1147extern void __tcp_put_md5sig_pool(void);
1148
1149static inline
1150struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1151{
1152 int cpu = get_cpu();
1153 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1154 if (!ret)
1155 put_cpu();
1156 return ret;
1157}
1158
1159static inline void tcp_put_md5sig_pool(void)
1160{
1161 __tcp_put_md5sig_pool();
1162 put_cpu();
1163}
1164
David S. Millerfe067e82007-03-07 12:12:44 -08001165/* write queue abstraction */
1166static inline void tcp_write_queue_purge(struct sock *sk)
1167{
1168 struct sk_buff *skb;
1169
1170 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1171 sk_stream_free_skb(sk, skb);
1172 sk_stream_mem_reclaim(sk);
1173}
1174
1175static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1176{
1177 struct sk_buff *skb = sk->sk_write_queue.next;
1178 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1179 return NULL;
1180 return skb;
1181}
1182
1183static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1184{
1185 struct sk_buff *skb = sk->sk_write_queue.prev;
1186 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1187 return NULL;
1188 return skb;
1189}
1190
1191static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1192{
1193 return skb->next;
1194}
1195
1196#define tcp_for_write_queue(skb, sk) \
1197 for (skb = (sk)->sk_write_queue.next; \
1198 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1199 skb = skb->next)
1200
1201#define tcp_for_write_queue_from(skb, sk) \
1202 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1203 skb = skb->next)
1204
1205static inline struct sk_buff *tcp_send_head(struct sock *sk)
1206{
1207 return sk->sk_send_head;
1208}
1209
1210static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1211{
1212 sk->sk_send_head = skb->next;
1213 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1214 sk->sk_send_head = NULL;
1215}
1216
1217static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1218{
1219 if (sk->sk_send_head == skb_unlinked)
1220 sk->sk_send_head = NULL;
1221}
1222
1223static inline void tcp_init_send_head(struct sock *sk)
1224{
1225 sk->sk_send_head = NULL;
1226}
1227
1228static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1229{
1230 __skb_queue_tail(&sk->sk_write_queue, skb);
1231}
1232
1233static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1234{
1235 __tcp_add_write_queue_tail(sk, skb);
1236
1237 /* Queue it, remembering where we must start sending. */
1238 if (sk->sk_send_head == NULL)
1239 sk->sk_send_head = skb;
1240}
1241
1242static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1243{
1244 __skb_queue_head(&sk->sk_write_queue, skb);
1245}
1246
1247/* Insert buff after skb on the write queue of sk. */
1248static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1249 struct sk_buff *buff,
1250 struct sock *sk)
1251{
1252 __skb_append(skb, buff, &sk->sk_write_queue);
1253}
1254
1255/* Insert skb between prev and next on the write queue of sk. */
1256static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1257 struct sk_buff *skb,
1258 struct sock *sk)
1259{
1260 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1261}
1262
1263static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1264{
1265 __skb_unlink(skb, &sk->sk_write_queue);
1266}
1267
1268static inline int tcp_skb_is_last(const struct sock *sk,
1269 const struct sk_buff *skb)
1270{
1271 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1272}
1273
1274static inline int tcp_write_queue_empty(struct sock *sk)
1275{
1276 return skb_queue_empty(&sk->sk_write_queue);
1277}
1278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279/* /proc */
1280enum tcp_seq_states {
1281 TCP_SEQ_STATE_LISTENING,
1282 TCP_SEQ_STATE_OPENREQ,
1283 TCP_SEQ_STATE_ESTABLISHED,
1284 TCP_SEQ_STATE_TIME_WAIT,
1285};
1286
1287struct tcp_seq_afinfo {
1288 struct module *owner;
1289 char *name;
1290 sa_family_t family;
1291 int (*seq_show) (struct seq_file *m, void *v);
1292 struct file_operations *seq_fops;
1293};
1294
1295struct tcp_iter_state {
1296 sa_family_t family;
1297 enum tcp_seq_states state;
1298 struct sock *syn_wait_sk;
1299 int bucket, sbucket, num, uid;
1300 struct seq_operations seq_ops;
1301};
1302
1303extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1304extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1305
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001306extern struct request_sock_ops tcp_request_sock_ops;
1307
1308extern int tcp_v4_destroy_sock(struct sock *sk);
1309
Herbert Xua430a432006-07-08 13:34:56 -07001310extern int tcp_v4_gso_send_check(struct sk_buff *skb);
Herbert Xu576a30e2006-06-27 13:22:38 -07001311extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001312
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001313#ifdef CONFIG_PROC_FS
1314extern int tcp4_proc_init(void);
1315extern void tcp4_proc_exit(void);
1316#endif
1317
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001318/* TCP af-specific functions */
1319struct tcp_sock_af_ops {
1320#ifdef CONFIG_TCP_MD5SIG
1321 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1322 struct sock *addr_sk);
1323 int (*calc_md5_hash) (char *location,
1324 struct tcp_md5sig_key *md5,
1325 struct sock *sk,
1326 struct dst_entry *dst,
1327 struct request_sock *req,
1328 struct tcphdr *th,
1329 int protocol, int len);
1330 int (*md5_add) (struct sock *sk,
1331 struct sock *addr_sk,
1332 u8 *newkey,
1333 u8 len);
1334 int (*md5_parse) (struct sock *sk,
1335 char __user *optval,
1336 int optlen);
1337#endif
1338};
1339
1340struct tcp_request_sock_ops {
1341#ifdef CONFIG_TCP_MD5SIG
1342 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1343 struct request_sock *req);
1344#endif
1345};
1346
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001347extern void tcp_v4_init(struct net_proto_family *ops);
1348extern void tcp_init(void);
1349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350#endif /* _TCP_H */