blob: e22b4f0305a339ce402b1e959ee0a64665fe2b1e [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
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700423extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
424 int nonagle);
425extern int tcp_may_send_now(struct sock *sk);
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
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700482static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700484 struct tcp_sock *tp = tcp_sk(sk);
485
David S. Millerb03efcf2005-07-08 14:57:23 -0700486 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 tp->rcv_wnd &&
488 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
489 !tp->urg_data)
490 tcp_fast_path_on(tp);
491}
492
493/* Compute the actual receive window we are currently advertising.
494 * Rcv_nxt can be after the window if our peer push more data
495 * than the offered window.
496 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800497static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
499 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
500
501 if (win < 0)
502 win = 0;
503 return (u32) win;
504}
505
506/* Choose a new window, without checks for shrinking, and without
507 * scaling applied to the result. The caller does these things
508 * if necessary. This is a "raw" window selection.
509 */
510extern u32 __tcp_select_window(struct sock *sk);
511
512/* TCP timestamps are only 32-bits, this causes a slight
513 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800514 * of jiffies in the buffer control blocks below. We decided
515 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 * casts with the following macro.
517 */
518#define tcp_time_stamp ((__u32)(jiffies))
519
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800520/* This is what the send packet queuing engine uses to pass
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 * TCP per-packet control information to the transmission
522 * code. We also store the host-order sequence numbers in
523 * here too. This is 36 bytes on 32-bit architectures,
524 * 40 bytes on 64-bit machines, if this grows please adjust
525 * skbuff.h:skbuff->cb[xxx] size appropriately.
526 */
527struct tcp_skb_cb {
528 union {
529 struct inet_skb_parm h4;
530#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
531 struct inet6_skb_parm h6;
532#endif
533 } header; /* For incoming frames */
534 __u32 seq; /* Starting sequence number */
535 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
536 __u32 when; /* used to compute rtt's */
537 __u8 flags; /* TCP header flags. */
538
539 /* NOTE: These must match up to the flags byte in a
540 * real TCP header.
541 */
542#define TCPCB_FLAG_FIN 0x01
543#define TCPCB_FLAG_SYN 0x02
544#define TCPCB_FLAG_RST 0x04
545#define TCPCB_FLAG_PSH 0x08
546#define TCPCB_FLAG_ACK 0x10
547#define TCPCB_FLAG_URG 0x20
548#define TCPCB_FLAG_ECE 0x40
549#define TCPCB_FLAG_CWR 0x80
550
551 __u8 sacked; /* State flags for SACK/FACK. */
552#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
553#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
554#define TCPCB_LOST 0x04 /* SKB is lost */
555#define TCPCB_TAGBITS 0x07 /* All tag bits */
556
557#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
558#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
559
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800560#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562#define TCPCB_AT_TAIL (TCPCB_URG)
563
564 __u16 urg_ptr; /* Valid w/URG flags is set. */
565 __u32 ack_seq; /* Sequence number ACK'd */
566};
567
568#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
569
570#include <net/tcp_ecn.h>
571
572/* Due to TSO, an SKB can be composed of multiple actual
573 * packets. To keep these tracked properly, we use this.
574 */
575static inline int tcp_skb_pcount(const struct sk_buff *skb)
576{
Herbert Xu79671682006-06-22 02:40:14 -0700577 return skb_shinfo(skb)->gso_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
580/* This is valid iff tcp_skb_pcount() > 1. */
581static inline int tcp_skb_mss(const struct sk_buff *skb)
582{
Herbert Xu79671682006-06-22 02:40:14 -0700583 return skb_shinfo(skb)->gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
586static inline void tcp_dec_pcount_approx(__u32 *count,
587 const struct sk_buff *skb)
588{
589 if (*count) {
590 *count -= tcp_skb_pcount(skb);
591 if ((int)*count < 0)
592 *count = 0;
593 }
594}
595
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700596static inline void tcp_packets_out_inc(struct sock *sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 const struct sk_buff *skb)
598{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700599 struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 int orig = tp->packets_out;
601
602 tp->packets_out += tcp_skb_pcount(skb);
603 if (!orig)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700604 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
605 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606}
607
608static inline void tcp_packets_out_dec(struct tcp_sock *tp,
609 const struct sk_buff *skb)
610{
611 tp->packets_out -= tcp_skb_pcount(skb);
612}
613
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700614/* Events passed to congestion control interface */
615enum tcp_ca_event {
616 CA_EVENT_TX_START, /* first transmit when no packets in flight */
617 CA_EVENT_CWND_RESTART, /* congestion window restart */
618 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
619 CA_EVENT_FRTO, /* fast recovery timeout */
620 CA_EVENT_LOSS, /* loss timeout */
621 CA_EVENT_FAST_ACK, /* in sequence ack */
622 CA_EVENT_SLOW_ACK, /* other ack */
623};
624
625/*
626 * Interface for adding new TCP congestion control handlers
627 */
628#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800629#define TCP_CA_MAX 128
630#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
631
Stephen Hemminger164891a2007-04-23 22:26:16 -0700632#define TCP_CONG_NON_RESTRICTED 0x1
633#define TCP_CONG_RTT_STAMP 0x2
634
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700635struct tcp_congestion_ops {
636 struct list_head list;
Stephen Hemminger164891a2007-04-23 22:26:16 -0700637 unsigned long flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700638
639 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300640 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700641 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300642 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700643
644 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300645 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700646 /* lower bound for congestion window (optional) */
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700647 u32 (*min_cwnd)(const struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700648 /* do new cwnd calculation (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300649 void (*cong_avoid)(struct sock *sk, u32 ack,
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700650 u32 rtt, u32 in_flight, int good_ack);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700651 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300652 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700653 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300654 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700655 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300656 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700657 /* hook for packet ack accounting (optional) */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700658 void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300659 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300660 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700661
662 char name[TCP_CA_NAME_MAX];
663 struct module *owner;
664};
665
666extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
667extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
668
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300669extern void tcp_init_congestion_control(struct sock *sk);
670extern void tcp_cleanup_congestion_control(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700671extern int tcp_set_default_congestion_control(const char *name);
672extern void tcp_get_default_congestion_control(char *name);
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800673extern void tcp_get_available_congestion_control(char *buf, size_t len);
Stephen Hemmingerce7bc3b2006-11-09 16:35:15 -0800674extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
675extern int tcp_set_allowed_congestion_control(char *allowed);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300676extern int tcp_set_congestion_control(struct sock *sk, const char *name);
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800677extern void tcp_slow_start(struct tcp_sock *tp);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700678
Stephen Hemminger5f8ef482005-06-23 20:37:36 -0700679extern struct tcp_congestion_ops tcp_init_congestion_ops;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300680extern u32 tcp_reno_ssthresh(struct sock *sk);
681extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700682 u32 rtt, u32 in_flight, int flag);
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700683extern u32 tcp_reno_min_cwnd(const struct sock *sk);
David S. Millera8acfba2005-06-23 23:45:02 -0700684extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700685
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300686static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700687{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300688 struct inet_connection_sock *icsk = inet_csk(sk);
689
690 if (icsk->icsk_ca_ops->set_state)
691 icsk->icsk_ca_ops->set_state(sk, ca_state);
692 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700693}
694
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300695static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700696{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300697 const struct inet_connection_sock *icsk = inet_csk(sk);
698
699 if (icsk->icsk_ca_ops->cwnd_event)
700 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700701}
702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703/* This determines how many packets are "in the network" to the best
704 * of our knowledge. In many cases it is conservative, but where
705 * detailed information is available from the receiver (via SACK
706 * blocks etc.) we can make more aggressive calculations.
707 *
708 * Use this for decisions involving congestion control, use just
709 * tp->packets_out to determine if the send queue is empty or not.
710 *
711 * Read this equation as:
712 *
713 * "Packets sent once on transmission queue" MINUS
714 * "Packets left network, but not honestly ACKed yet" PLUS
715 * "Packets fast retransmitted"
716 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800717static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
719 return (tp->packets_out - tp->left_out + tp->retrans_out);
720}
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
723 * The exception is rate halving phase, when cwnd is decreasing towards
724 * ssthresh.
725 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300726static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300728 const struct tcp_sock *tp = tcp_sk(sk);
729 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 return tp->snd_ssthresh;
731 else
732 return max(tp->snd_ssthresh,
733 ((tp->snd_cwnd >> 1) +
734 (tp->snd_cwnd >> 2)));
735}
736
737static inline void tcp_sync_left_out(struct tcp_sock *tp)
738{
Ilpo Järvinen0ec96822007-05-03 03:30:34 -0700739 BUG_ON(tp->rx_opt.sack_ok &&
740 (tp->sacked_out + tp->lost_out > tp->packets_out));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 tp->left_out = tp->sacked_out + tp->lost_out;
742}
743
Ilpo Järvinen3cfe3ba2007-02-27 10:09:49 -0800744extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
746
747/* Slow start with delack produces 3 packets of burst, so that
748 * it is safe "de facto".
749 */
750static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
751{
752 return 3;
753}
754
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800755/* RFC2861 Check whether we are limited by application or congestion window
756 * This is the inverse of cwnd check in tcp_tso_should_defer
757 */
758static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
759{
760 const struct tcp_sock *tp = tcp_sk(sk);
761 u32 left;
762
763 if (in_flight >= tp->snd_cwnd)
764 return 1;
765
Herbert Xubcd76112006-06-30 13:36:35 -0700766 if (!sk_can_gso(sk))
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800767 return 0;
768
769 left = tp->snd_cwnd - in_flight;
770 if (sysctl_tcp_tso_win_divisor)
771 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
772 else
773 return left <= tcp_max_burst(tp);
774}
775
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800776static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
777 const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
779 if (skb->len < mss)
780 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
781}
782
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700783static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700785 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700786 const struct inet_connection_sock *icsk = inet_csk(sk);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700787
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700788 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700789 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
790 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700793static inline void tcp_push_pending_frames(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700795 struct tcp_sock *tp = tcp_sk(sk);
796
797 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798}
799
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800800static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
802 tp->snd_wl1 = seq;
803}
804
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800805static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
807 tp->snd_wl1 = seq;
808}
809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810/*
811 * Calculate(/check) TCP checksum
812 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800813static inline __sum16 tcp_v4_check(int len, __be32 saddr,
814 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815{
816 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
817}
818
Al Virob51655b2006-11-14 21:40:42 -0800819static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
Herbert Xufb286bb2005-11-10 13:01:24 -0800821 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}
823
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800824static inline int tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
Herbert Xu60476372007-04-09 11:59:39 -0700826 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 __tcp_checksum_complete(skb);
828}
829
830/* Prequeue for VJ style copy to user, combined with checksumming. */
831
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800832static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 tp->ucopy.task = NULL;
835 tp->ucopy.len = 0;
836 tp->ucopy.memory = 0;
837 skb_queue_head_init(&tp->ucopy.prequeue);
Chris Leech97fc2f02006-05-23 17:55:33 -0700838#ifdef CONFIG_NET_DMA
839 tp->ucopy.dma_chan = NULL;
840 tp->ucopy.wakeup = 0;
841 tp->ucopy.pinned_list = NULL;
842 tp->ucopy.dma_cookie = 0;
843#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844}
845
846/* Packet is added to VJ-style prequeue for processing in process
847 * context, if a reader task is waiting. Apparently, this exciting
848 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
849 * failed somewhere. Latency? Burstiness? Well, at least now we will
850 * see, why it failed. 8)8) --ANK
851 *
852 * NOTE: is this not too big to inline?
853 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800854static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855{
856 struct tcp_sock *tp = tcp_sk(sk);
857
858 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
859 __skb_queue_tail(&tp->ucopy.prequeue, skb);
860 tp->ucopy.memory += skb->truesize;
861 if (tp->ucopy.memory > sk->sk_rcvbuf) {
862 struct sk_buff *skb1;
863
864 BUG_ON(sock_owned_by_user(sk));
865
866 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
867 sk->sk_backlog_rcv(sk, skb1);
868 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
869 }
870
871 tp->ucopy.memory = 0;
872 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
873 wake_up_interruptible(sk->sk_sleep);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700874 if (!inet_csk_ack_scheduled(sk))
875 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700876 (3 * TCP_RTO_MIN) / 4,
877 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879 return 1;
880 }
881 return 0;
882}
883
884
885#undef STATE_TRACE
886
887#ifdef STATE_TRACE
888static const char *statename[]={
889 "Unused","Established","Syn Sent","Syn Recv",
890 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
891 "Close Wait","Last ACK","Listen","Closing"
892};
893#endif
894
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800895static inline void tcp_set_state(struct sock *sk, int state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
897 int oldstate = sk->sk_state;
898
899 switch (state) {
900 case TCP_ESTABLISHED:
901 if (oldstate != TCP_ESTABLISHED)
902 TCP_INC_STATS(TCP_MIB_CURRESTAB);
903 break;
904
905 case TCP_CLOSE:
906 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
907 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
908
909 sk->sk_prot->unhash(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700910 if (inet_csk(sk)->icsk_bind_hash &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
Arnaldo Carvalho de Melo2d8c4ce2005-08-09 20:07:13 -0700912 inet_put_port(&tcp_hashinfo, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 /* fall through */
914 default:
915 if (oldstate==TCP_ESTABLISHED)
916 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
917 }
918
919 /* Change state AFTER socket is unhashed to avoid closed
920 * socket sitting in hash tables.
921 */
922 sk->sk_state = state;
923
924#ifdef STATE_TRACE
925 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
926#endif
927}
928
Andi Kleen4ac02ba2007-04-20 17:11:46 -0700929extern void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800931static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932{
933 rx_opt->dsack = 0;
934 rx_opt->eff_sacks = 0;
935 rx_opt->num_sacks = 0;
936}
937
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938/* Determine a window scaling and initial window to offer. */
939extern void tcp_select_initial_window(int __space, __u32 mss,
940 __u32 *rcv_wnd, __u32 *window_clamp,
941 int wscale_ok, __u8 *rcv_wscale);
942
943static inline int tcp_win_from_space(int space)
944{
945 return sysctl_tcp_adv_win_scale<=0 ?
946 (space>>(-sysctl_tcp_adv_win_scale)) :
947 space - (space>>sysctl_tcp_adv_win_scale);
948}
949
950/* Note: caller must be prepared to deal with negative returns */
951static inline int tcp_space(const struct sock *sk)
952{
953 return tcp_win_from_space(sk->sk_rcvbuf -
954 atomic_read(&sk->sk_rmem_alloc));
955}
956
957static inline int tcp_full_space(const struct sock *sk)
958{
959 return tcp_win_from_space(sk->sk_rcvbuf);
960}
961
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800962static inline void tcp_openreq_init(struct request_sock *req,
963 struct tcp_options_received *rx_opt,
964 struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700966 struct inet_request_sock *ireq = inet_rsk(req);
967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700969 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 req->mss = rx_opt->mss_clamp;
971 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700972 ireq->tstamp_ok = rx_opt->tstamp_ok;
973 ireq->sack_ok = rx_opt->sack_ok;
974 ireq->snd_wscale = rx_opt->snd_wscale;
975 ireq->wscale_ok = rx_opt->wscale_ok;
976 ireq->acked = 0;
977 ireq->ecn_ok = 0;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700978 ireq->rmt_port = tcp_hdr(skb)->source;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979}
980
981extern void tcp_enter_memory_pressure(void);
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983static inline int keepalive_intvl_when(const struct tcp_sock *tp)
984{
985 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
986}
987
988static inline int keepalive_time_when(const struct tcp_sock *tp)
989{
990 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
991}
992
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700993static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700995 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
996 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700998 if (fin_timeout < (rto << 2) - (rto >> 1))
999 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
1001 return fin_timeout;
1002}
1003
1004static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1005{
1006 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1007 return 0;
James Morris9d729f72007-03-04 16:12:44 -08001008 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 return 0;
1010
1011 /* RST segments are not recommended to carry timestamp,
1012 and, if they do, it is recommended to ignore PAWS because
1013 "their cleanup function should take precedence over timestamps."
1014 Certainly, it is mistake. It is necessary to understand the reasons
1015 of this constraint to relax it: if peer reboots, clock may go
1016 out-of-sync and half-open connections will not be reset.
1017 Actually, the problem would be not existing if all
1018 the implementations followed draft about maintaining clock
1019 via reboots. Linux-2.2 DOES NOT!
1020
1021 However, we can relax time bounds for RST segments to MSL.
1022 */
James Morris9d729f72007-03-04 16:12:44 -08001023 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 return 0;
1025 return 1;
1026}
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028#define TCP_CHECK_TIMER(sk) do { } while (0)
1029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030static inline void tcp_mib_init(void)
1031{
1032 /* See RFC 2012 */
1033 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1034 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1035 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1036 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1037}
1038
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001039/*from STCP */
1040static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1041 tp->lost_skb_hint = NULL;
1042 tp->scoreboard_skb_hint = NULL;
1043 tp->retransmit_skb_hint = NULL;
1044 tp->forward_skb_hint = NULL;
1045 tp->fastpath_skb_hint = NULL;
1046}
1047
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001048/* MD5 Signature */
1049struct crypto_hash;
1050
1051/* - key database */
1052struct tcp_md5sig_key {
1053 u8 *key;
1054 u8 keylen;
1055};
1056
1057struct tcp4_md5sig_key {
1058 u8 *key;
1059 u16 keylen;
1060 __be32 addr;
1061};
1062
1063struct tcp6_md5sig_key {
1064 u8 *key;
1065 u16 keylen;
1066#if 0
1067 u32 scope_id; /* XXX */
1068#endif
1069 struct in6_addr addr;
1070};
1071
1072/* - sock block */
1073struct tcp_md5sig_info {
1074 struct tcp4_md5sig_key *keys4;
1075#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1076 struct tcp6_md5sig_key *keys6;
1077 u32 entries6;
1078 u32 alloced6;
1079#endif
1080 u32 entries4;
1081 u32 alloced4;
1082};
1083
1084/* - pseudo header */
1085struct tcp4_pseudohdr {
1086 __be32 saddr;
1087 __be32 daddr;
1088 __u8 pad;
1089 __u8 protocol;
1090 __be16 len;
1091};
1092
1093struct tcp6_pseudohdr {
1094 struct in6_addr saddr;
1095 struct in6_addr daddr;
1096 __be32 len;
1097 __be32 protocol; /* including padding */
1098};
1099
1100union tcp_md5sum_block {
1101 struct tcp4_pseudohdr ip4;
1102#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1103 struct tcp6_pseudohdr ip6;
1104#endif
1105};
1106
1107/* - pool: digest algorithm, hash description and scratch buffer */
1108struct tcp_md5sig_pool {
1109 struct hash_desc md5_desc;
1110 union tcp_md5sum_block md5_blk;
1111};
1112
1113#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1114
1115/* - functions */
1116extern int tcp_v4_calc_md5_hash(char *md5_hash,
1117 struct tcp_md5sig_key *key,
1118 struct sock *sk,
1119 struct dst_entry *dst,
1120 struct request_sock *req,
1121 struct tcphdr *th,
1122 int protocol, int tcplen);
1123extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1124 struct sock *addr_sk);
1125
1126extern int tcp_v4_md5_do_add(struct sock *sk,
1127 __be32 addr,
1128 u8 *newkey,
1129 u8 newkeylen);
1130
1131extern int tcp_v4_md5_do_del(struct sock *sk,
Al Viro8e5200f2006-11-20 18:06:37 -08001132 __be32 addr);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001133
1134extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1135extern void tcp_free_md5sig_pool(void);
1136
1137extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1138extern void __tcp_put_md5sig_pool(void);
1139
1140static inline
1141struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1142{
1143 int cpu = get_cpu();
1144 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1145 if (!ret)
1146 put_cpu();
1147 return ret;
1148}
1149
1150static inline void tcp_put_md5sig_pool(void)
1151{
1152 __tcp_put_md5sig_pool();
1153 put_cpu();
1154}
1155
David S. Millerfe067e82007-03-07 12:12:44 -08001156/* write queue abstraction */
1157static inline void tcp_write_queue_purge(struct sock *sk)
1158{
1159 struct sk_buff *skb;
1160
1161 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1162 sk_stream_free_skb(sk, skb);
1163 sk_stream_mem_reclaim(sk);
1164}
1165
1166static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1167{
1168 struct sk_buff *skb = sk->sk_write_queue.next;
1169 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1170 return NULL;
1171 return skb;
1172}
1173
1174static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1175{
1176 struct sk_buff *skb = sk->sk_write_queue.prev;
1177 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1178 return NULL;
1179 return skb;
1180}
1181
1182static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1183{
1184 return skb->next;
1185}
1186
1187#define tcp_for_write_queue(skb, sk) \
1188 for (skb = (sk)->sk_write_queue.next; \
1189 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1190 skb = skb->next)
1191
1192#define tcp_for_write_queue_from(skb, sk) \
1193 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1194 skb = skb->next)
1195
1196static inline struct sk_buff *tcp_send_head(struct sock *sk)
1197{
1198 return sk->sk_send_head;
1199}
1200
1201static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1202{
Ilpo Järvinend551e452007-04-30 00:42:20 -07001203 struct tcp_sock *tp = tcp_sk(sk);
1204
David S. Millerfe067e82007-03-07 12:12:44 -08001205 sk->sk_send_head = skb->next;
1206 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1207 sk->sk_send_head = NULL;
Ilpo Järvinend551e452007-04-30 00:42:20 -07001208 /* Don't override Nagle indefinately with F-RTO */
1209 if (tp->frto_counter == 2)
1210 tp->frto_counter = 3;
David S. Millerfe067e82007-03-07 12:12:44 -08001211}
1212
1213static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1214{
1215 if (sk->sk_send_head == skb_unlinked)
1216 sk->sk_send_head = NULL;
1217}
1218
1219static inline void tcp_init_send_head(struct sock *sk)
1220{
1221 sk->sk_send_head = NULL;
1222}
1223
1224static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1225{
1226 __skb_queue_tail(&sk->sk_write_queue, skb);
1227}
1228
1229static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1230{
1231 __tcp_add_write_queue_tail(sk, skb);
1232
1233 /* Queue it, remembering where we must start sending. */
1234 if (sk->sk_send_head == NULL)
1235 sk->sk_send_head = skb;
1236}
1237
1238static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1239{
1240 __skb_queue_head(&sk->sk_write_queue, skb);
1241}
1242
1243/* Insert buff after skb on the write queue of sk. */
1244static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1245 struct sk_buff *buff,
1246 struct sock *sk)
1247{
1248 __skb_append(skb, buff, &sk->sk_write_queue);
1249}
1250
1251/* Insert skb between prev and next on the write queue of sk. */
1252static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1253 struct sk_buff *skb,
1254 struct sock *sk)
1255{
1256 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1257}
1258
1259static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1260{
1261 __skb_unlink(skb, &sk->sk_write_queue);
1262}
1263
1264static inline int tcp_skb_is_last(const struct sock *sk,
1265 const struct sk_buff *skb)
1266{
1267 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1268}
1269
1270static inline int tcp_write_queue_empty(struct sock *sk)
1271{
1272 return skb_queue_empty(&sk->sk_write_queue);
1273}
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275/* /proc */
1276enum tcp_seq_states {
1277 TCP_SEQ_STATE_LISTENING,
1278 TCP_SEQ_STATE_OPENREQ,
1279 TCP_SEQ_STATE_ESTABLISHED,
1280 TCP_SEQ_STATE_TIME_WAIT,
1281};
1282
1283struct tcp_seq_afinfo {
1284 struct module *owner;
1285 char *name;
1286 sa_family_t family;
1287 int (*seq_show) (struct seq_file *m, void *v);
1288 struct file_operations *seq_fops;
1289};
1290
1291struct tcp_iter_state {
1292 sa_family_t family;
1293 enum tcp_seq_states state;
1294 struct sock *syn_wait_sk;
1295 int bucket, sbucket, num, uid;
1296 struct seq_operations seq_ops;
1297};
1298
1299extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1300extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1301
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001302extern struct request_sock_ops tcp_request_sock_ops;
1303
1304extern int tcp_v4_destroy_sock(struct sock *sk);
1305
Herbert Xua430a432006-07-08 13:34:56 -07001306extern int tcp_v4_gso_send_check(struct sk_buff *skb);
Herbert Xu576a30e2006-06-27 13:22:38 -07001307extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001308
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001309#ifdef CONFIG_PROC_FS
1310extern int tcp4_proc_init(void);
1311extern void tcp4_proc_exit(void);
1312#endif
1313
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001314/* TCP af-specific functions */
1315struct tcp_sock_af_ops {
1316#ifdef CONFIG_TCP_MD5SIG
1317 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1318 struct sock *addr_sk);
1319 int (*calc_md5_hash) (char *location,
1320 struct tcp_md5sig_key *md5,
1321 struct sock *sk,
1322 struct dst_entry *dst,
1323 struct request_sock *req,
1324 struct tcphdr *th,
1325 int protocol, int len);
1326 int (*md5_add) (struct sock *sk,
1327 struct sock *addr_sk,
1328 u8 *newkey,
1329 u8 len);
1330 int (*md5_parse) (struct sock *sk,
1331 char __user *optval,
1332 int optlen);
1333#endif
1334};
1335
1336struct tcp_request_sock_ops {
1337#ifdef CONFIG_TCP_MD5SIG
1338 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1339 struct request_sock *req);
1340#endif
1341};
1342
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001343extern void tcp_v4_init(struct net_proto_family *ops);
1344extern void tcp_init(void);
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346#endif /* _TCP_H */