blob: 3e4b33e36602caade361654d0561a65b3fba2224 [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>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080032#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000033#include <linux/kref.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070034
35#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070036#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070037#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070039#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <net/sock.h>
41#include <net/snmp.h>
42#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070043#include <net/tcp_states.h>
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -070044#include <net/inet_ecn.h>
Satoru SATOH0c266892009-05-04 11:11:01 -070045#include <net/dst.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/seq_file.h>
48
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070049extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Eric Dumazetdd24c002008-11-25 21:17:14 -080051extern struct percpu_counter tcp_orphan_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052extern void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070055#define MAX_TCP_OPTION_SPACE 40
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57/*
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
60 */
61#define MAX_TCP_WINDOW 32767U
62
63/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64#define TCP_MIN_MSS 88U
65
John Heffner5d424d52006-03-20 17:53:41 -080066/* The least MTU to use for probing */
67#define TCP_BASE_MSS 512
68
Linus Torvalds1da177e2005-04-16 15:20:36 -070069/* After receiving this amount of duplicate ACKs fast retransmit starts. */
70#define TCP_FASTRETRANS_THRESH 3
71
72/* Maximal reordering. */
73#define TCP_MAX_REORDERING 127
74
75/* Maximal number of ACKs sent quickly to accelerate slow-start. */
76#define TCP_MAX_QUICKACKS 16U
77
78/* urg_data states */
79#define TCP_URG_VALID 0x0100
80#define TCP_URG_NOTYET 0x0200
81#define TCP_URG_READ 0x0400
82
83#define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
88 */
89
90#define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
95 */
96
97#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080098 * connection: ~180sec is RFC minimum */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800101 * connection: ~180sec is RFC minimum */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103
104#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
106 */
107
108
109#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
115 * TIME-WAIT timer.
116 */
117
118#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
119#if HZ >= 100
120#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121#define TCP_ATO_MIN ((unsigned)(HZ/25))
122#else
123#define TCP_DELACK_MIN 4U
124#define TCP_ATO_MIN 4U
125#endif
126#define TCP_RTO_MAX ((unsigned)(120*HZ))
127#define TCP_RTO_MIN ((unsigned)(HZ/5))
128#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
129
130#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
132 */
133
134#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136#define TCP_KEEPALIVE_INTVL (75*HZ)
137
138#define MAX_TCP_KEEPIDLE 32767
139#define MAX_TCP_KEEPINTVL 32767
140#define MAX_TCP_KEEPCNT 127
141#define MAX_TCP_SYNCNT 127
142
143#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
151 */
152#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156/*
157 * TCP option
158 */
159
160#define TCPOPT_NOP 1 /* Padding */
161#define TCPOPT_EOL 0 /* End of options */
162#define TCPOPT_MSS 2 /* Segment size negotiating */
163#define TCPOPT_WINDOW 3 /* Window scaling */
164#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165#define TCPOPT_SACK 5 /* SACK Block */
166#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800167#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
William Allen Simpson435cf552009-12-02 18:17:05 +0000168#define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170/*
171 * TCP option lengths
172 */
173
174#define TCPOLEN_MSS 4
175#define TCPOLEN_WINDOW 3
176#define TCPOLEN_SACK_PERM 2
177#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800178#define TCPOLEN_MD5SIG 18
William Allen Simpson435cf552009-12-02 18:17:05 +0000179#define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
180#define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
181#define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
182#define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184/* But this is what stacks really send out. */
185#define TCPOLEN_TSTAMP_ALIGNED 12
186#define TCPOLEN_WSCALE_ALIGNED 4
187#define TCPOLEN_SACKPERM_ALIGNED 4
188#define TCPOLEN_SACK_BASE 2
189#define TCPOLEN_SACK_BASE_ALIGNED 4
190#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800191#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700192#define TCPOLEN_MSS_ALIGNED 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194/* Flags in tp->nonagle */
195#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
196#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800197#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000199/* TCP thin-stream limits */
200#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
201
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700202extern struct inet_timewait_death_row tcp_death_row;
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205extern int sysctl_tcp_timestamps;
206extern int sysctl_tcp_window_scaling;
207extern int sysctl_tcp_sack;
208extern int sysctl_tcp_fin_timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209extern int sysctl_tcp_keepalive_time;
210extern int sysctl_tcp_keepalive_probes;
211extern int sysctl_tcp_keepalive_intvl;
212extern int sysctl_tcp_syn_retries;
213extern int sysctl_tcp_synack_retries;
214extern int sysctl_tcp_retries1;
215extern int sysctl_tcp_retries2;
216extern int sysctl_tcp_orphan_retries;
217extern int sysctl_tcp_syncookies;
218extern int sysctl_tcp_retrans_collapse;
219extern int sysctl_tcp_stdurg;
220extern int sysctl_tcp_rfc1337;
221extern int sysctl_tcp_abort_on_overflow;
222extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223extern int sysctl_tcp_fack;
224extern int sysctl_tcp_reordering;
225extern int sysctl_tcp_ecn;
226extern int sysctl_tcp_dsack;
227extern int sysctl_tcp_mem[3];
228extern int sysctl_tcp_wmem[3];
229extern int sysctl_tcp_rmem[3];
230extern int sysctl_tcp_app_win;
231extern int sysctl_tcp_adv_win_scale;
232extern int sysctl_tcp_tw_reuse;
233extern int sysctl_tcp_frto;
Ilpo Järvinen3cfe3ba2007-02-27 10:09:49 -0800234extern int sysctl_tcp_frto_response;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235extern int sysctl_tcp_low_latency;
Chris Leech95937822006-05-23 18:02:55 -0700236extern int sysctl_tcp_dma_copybreak;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238extern int sysctl_tcp_moderate_rcvbuf;
239extern int sysctl_tcp_tso_win_divisor;
Stephen Hemminger9772efb2005-11-10 17:09:53 -0800240extern int sysctl_tcp_abc;
John Heffner5d424d52006-03-20 17:53:41 -0800241extern int sysctl_tcp_mtu_probing;
242extern int sysctl_tcp_base_mss;
Rick Jones15d99e02006-03-20 22:40:29 -0800243extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700244extern int sysctl_tcp_slow_start_after_idle;
John Heffner886236c2007-03-25 19:21:45 -0700245extern int sysctl_tcp_max_ssthresh;
William Allen Simpson519855c2009-12-02 18:14:19 +0000246extern int sysctl_tcp_cookie_size;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000247extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000248extern int sysctl_tcp_thin_dupack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250extern atomic_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800251extern struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252extern int tcp_memory_pressure;
253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 * The next routines deal with comparing 32 bit unsigned ints
256 * and worry about wraparound (automatic with unsigned arithmetic).
257 */
258
259static inline int before(__u32 seq1, __u32 seq2)
260{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800261 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800263#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
265/* is s2<=s1<=s3 ? */
266static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
267{
268 return seq3 - seq2 >= seq1 - seq2;
269}
270
David S. Millerad1af0f2010-08-25 02:27:49 -0700271static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700272{
David S. Millerad1af0f2010-08-25 02:27:49 -0700273 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
274 int orphans = percpu_counter_read_positive(ocp);
275
276 if (orphans << shift > sysctl_tcp_max_orphans) {
277 orphans = percpu_counter_sum_positive(ocp);
278 if (orphans << shift > sysctl_tcp_max_orphans)
279 return true;
280 }
281
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
284 return true;
285 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700286}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Florian Westphala0f82f62009-04-19 09:43:48 +0000288/* syncookies: remember time of last synqueue overflow */
289static inline void tcp_synq_overflow(struct sock *sk)
290{
291 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
292}
293
294/* syncookies: no recent synqueue overflow on this listening socket? */
295static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
296{
297 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
298 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
299}
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301extern struct proto tcp_prot;
302
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700303#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
304#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
305#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
306#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000307#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Changli Gao53d31762010-07-10 20:41:06 +0000309extern void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310
Changli Gao53d31762010-07-10 20:41:06 +0000311extern void tcp_shutdown (struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Changli Gao53d31762010-07-10 20:41:06 +0000313extern int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
Changli Gao53d31762010-07-10 20:41:06 +0000315extern int tcp_v4_remember_stamp(struct sock *sk);
316extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Changli Gao7ba42912010-07-10 20:41:55 +0000317extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
318 size_t size);
319extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
320 size_t size, int flags);
Changli Gao53d31762010-07-10 20:41:06 +0000321extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
322extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
323 struct tcphdr *th, unsigned len);
324extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
325 struct tcphdr *th, unsigned len);
326extern void tcp_rcv_space_adjust(struct sock *sk);
327extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
328extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
329extern void tcp_twsk_destructor(struct sock *sk);
330extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
331 struct pipe_inode_info *pipe, size_t len,
332 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800333
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700334static inline void tcp_dec_quickack_mode(struct sock *sk,
335 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700337 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700338
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700339 if (icsk->icsk_ack.quick) {
340 if (pkts >= icsk->icsk_ack.quick) {
341 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700342 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700343 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700344 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700345 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 }
347}
348
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700349extern void tcp_enter_quickack_mode(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700351#define TCP_ECN_OK 1
352#define TCP_ECN_QUEUE_CWR 2
353#define TCP_ECN_DEMAND_CWR 4
354
355static __inline__ void
356TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
357{
358 if (sysctl_tcp_ecn && th->ece && th->cwr)
359 inet_rsk(req)->ecn_ok = 1;
360}
361
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000362enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 TCP_TW_SUCCESS = 0,
364 TCP_TW_RST = 1,
365 TCP_TW_ACK = 2,
366 TCP_TW_SYN = 3
367};
368
369
Changli Gao53d31762010-07-10 20:41:06 +0000370extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
371 struct sk_buff *skb,
372 const struct tcphdr *th);
373extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
374 struct request_sock *req,
375 struct request_sock **prev);
376extern int tcp_child_process(struct sock *parent, struct sock *child,
377 struct sk_buff *skb);
378extern int tcp_use_frto(struct sock *sk);
379extern void tcp_enter_frto(struct sock *sk);
380extern void tcp_enter_loss(struct sock *sk, int how);
381extern void tcp_clear_retrans(struct tcp_sock *tp);
382extern void tcp_update_metrics(struct sock *sk);
383extern void tcp_close(struct sock *sk, long timeout);
384extern unsigned int tcp_poll(struct file * file, struct socket *sock,
385 struct poll_table_struct *wait);
386extern int tcp_getsockopt(struct sock *sk, int level, int optname,
387 char __user *optval, int __user *optlen);
388extern int tcp_setsockopt(struct sock *sk, int level, int optname,
389 char __user *optval, unsigned int optlen);
390extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
391 char __user *optval, int __user *optlen);
392extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
393 char __user *optval, unsigned int optlen);
394extern void tcp_set_keepalive(struct sock *sk, int val);
395extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
396extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
397 size_t len, int nonblock, int flags, int *addr_len);
398extern void tcp_parse_options(struct sk_buff *skb,
399 struct tcp_options_received *opt_rx, u8 **hvpp,
400 int estab);
401extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403/*
404 * TCP v4 functions exported for the inet6 API
405 */
406
Changli Gao53d31762010-07-10 20:41:06 +0000407extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
408extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
409extern struct sock * tcp_create_openreq_child(struct sock *sk,
410 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 struct sk_buff *skb);
Changli Gao53d31762010-07-10 20:41:06 +0000412extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
413 struct request_sock *req,
414 struct dst_entry *dst);
415extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
416extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
417 int addr_len);
418extern int tcp_connect(struct sock *sk);
419extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
420 struct request_sock *req,
421 struct request_values *rvp);
422extern int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425/* From syncookies.c */
Florian Westphal2051f112008-03-23 22:21:28 -0700426extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
428 struct ip_options *opt);
429extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
430 __u16 *mss);
431
Florian Westphal4dfc2812008-04-10 03:12:40 -0700432extern __u32 cookie_init_timestamp(struct request_sock *req);
Florian Westphal172d69e2010-06-21 11:48:45 +0000433extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700434
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800435/* From net/ipv6/syncookies.c */
436extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
437extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
438 __u16 *mss);
439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440/* tcp_output.c */
441
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700442extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
443 int nonagle);
444extern int tcp_may_send_now(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000446extern void tcp_retransmit_timer(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447extern void tcp_xmit_retransmit_queue(struct sock *);
448extern void tcp_simple_retransmit(struct sock *);
449extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
David S. Miller6475be12005-09-01 22:47:01 -0700450extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452extern void tcp_send_probe0(struct sock *);
453extern void tcp_send_partial(struct sock *);
Changli Gao53d31762010-07-10 20:41:06 +0000454extern int tcp_write_wakeup(struct sock *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455extern void tcp_send_fin(struct sock *sk);
Al Virodd0fc662005-10-07 07:46:04 +0100456extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
Changli Gao53d31762010-07-10 20:41:06 +0000457extern int tcp_send_synack(struct sock *);
David S. Millerc1b4a7e2005-07-05 15:24:38 -0700458extern void tcp_push_one(struct sock *, unsigned int mss_now);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459extern void tcp_send_ack(struct sock *sk);
460extern void tcp_send_delayed_ack(struct sock *sk);
461
David S. Millera762a982005-07-05 15:18:51 -0700462/* tcp_input.c */
463extern void tcp_cwnd_application_limited(struct sock *sk);
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465/* tcp_timer.c */
466extern void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700467static inline void tcp_clear_xmit_timers(struct sock *sk)
468{
469 inet_csk_clear_xmit_timers(sk);
470}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000473extern unsigned int tcp_current_mss(struct sock *sk);
474
475/* Bound MSS / TSO packet size with the half of the window */
476static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
477{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700478 int cutoff;
479
480 /* When peer uses tiny windows, there is no use in packetizing
481 * to sub-MSS pieces for the sake of SWS or making sure there
482 * are enough packets in the pipe for fast recovery.
483 *
484 * On the other hand, for extremely large MSS devices, handling
485 * smaller than MSS windows in this way does make sense.
486 */
487 if (tp->max_window >= 512)
488 cutoff = (tp->max_window >> 1);
489 else
490 cutoff = tp->max_window;
491
492 if (cutoff && pktsize > cutoff)
493 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000494 else
495 return pktsize;
496}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300498/* tcp.c */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499extern void tcp_get_info(struct sock *, struct tcp_info *);
500
501/* Read 'sendfile()'-style from a TCP socket */
502typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
503 unsigned int, size_t);
504extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
505 sk_read_actor_t recv_actor);
506
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800507extern void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
John Heffner5d424d52006-03-20 17:53:41 -0800509extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
510extern int tcp_mss_to_mtu(struct sock *sk, int mss);
511extern void tcp_mtup_init(struct sock *sk);
512
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000513static inline void tcp_bound_rto(const struct sock *sk)
514{
515 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
516 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
517}
518
519static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
520{
521 return (tp->srtt >> 3) + tp->rttvar;
522}
523
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800524static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525{
526 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
527 ntohl(TCP_FLAG_ACK) |
528 snd_wnd);
529}
530
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800531static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532{
533 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
534}
535
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700536static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700538 struct tcp_sock *tp = tcp_sk(sk);
539
David S. Millerb03efcf2005-07-08 14:57:23 -0700540 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 tp->rcv_wnd &&
542 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
543 !tp->urg_data)
544 tcp_fast_path_on(tp);
545}
546
Satoru SATOH0c266892009-05-04 11:11:01 -0700547/* Compute the actual rto_min value */
548static inline u32 tcp_rto_min(struct sock *sk)
549{
550 struct dst_entry *dst = __sk_dst_get(sk);
551 u32 rto_min = TCP_RTO_MIN;
552
553 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
554 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
555 return rto_min;
556}
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558/* Compute the actual receive window we are currently advertising.
559 * Rcv_nxt can be after the window if our peer push more data
560 * than the offered window.
561 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800562static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
564 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
565
566 if (win < 0)
567 win = 0;
568 return (u32) win;
569}
570
571/* Choose a new window, without checks for shrinking, and without
572 * scaling applied to the result. The caller does these things
573 * if necessary. This is a "raw" window selection.
574 */
Changli Gao53d31762010-07-10 20:41:06 +0000575extern u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577/* TCP timestamps are only 32-bits, this causes a slight
578 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800579 * of jiffies in the buffer control blocks below. We decided
580 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * casts with the following macro.
582 */
583#define tcp_time_stamp ((__u32)(jiffies))
584
Changli Gaoa3433f32010-06-12 14:01:43 +0000585#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
586
587#define TCPHDR_FIN 0x01
588#define TCPHDR_SYN 0x02
589#define TCPHDR_RST 0x04
590#define TCPHDR_PSH 0x08
591#define TCPHDR_ACK 0x10
592#define TCPHDR_URG 0x20
593#define TCPHDR_ECE 0x40
594#define TCPHDR_CWR 0x80
595
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800596/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700597 * TCP per-packet control information to the transmission code.
598 * We also store the host-order sequence numbers in here too.
599 * This is 44 bytes if IPV6 is enabled.
600 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 */
602struct tcp_skb_cb {
603 union {
604 struct inet_skb_parm h4;
605#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
606 struct inet6_skb_parm h6;
607#endif
608 } header; /* For incoming frames */
609 __u32 seq; /* Starting sequence number */
610 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
611 __u32 when; /* used to compute rtt's */
612 __u8 flags; /* TCP header flags. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 __u8 sacked; /* State flags for SACK/FACK. */
614#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
615#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
616#define TCPCB_LOST 0x04 /* SKB is lost */
617#define TCPCB_TAGBITS 0x07 /* All tag bits */
618
619#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
620#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 __u32 ack_seq; /* Sequence number ACK'd */
623};
624
625#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627/* Due to TSO, an SKB can be composed of multiple actual
628 * packets. To keep these tracked properly, we use this.
629 */
630static inline int tcp_skb_pcount(const struct sk_buff *skb)
631{
Herbert Xu79671682006-06-22 02:40:14 -0700632 return skb_shinfo(skb)->gso_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
634
635/* This is valid iff tcp_skb_pcount() > 1. */
636static inline int tcp_skb_mss(const struct sk_buff *skb)
637{
Herbert Xu79671682006-06-22 02:40:14 -0700638 return skb_shinfo(skb)->gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639}
640
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700641/* Events passed to congestion control interface */
642enum tcp_ca_event {
643 CA_EVENT_TX_START, /* first transmit when no packets in flight */
644 CA_EVENT_CWND_RESTART, /* congestion window restart */
645 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
646 CA_EVENT_FRTO, /* fast recovery timeout */
647 CA_EVENT_LOSS, /* loss timeout */
648 CA_EVENT_FAST_ACK, /* in sequence ack */
649 CA_EVENT_SLOW_ACK, /* other ack */
650};
651
652/*
653 * Interface for adding new TCP congestion control handlers
654 */
655#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800656#define TCP_CA_MAX 128
657#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
658
Stephen Hemminger164891a2007-04-23 22:26:16 -0700659#define TCP_CONG_NON_RESTRICTED 0x1
660#define TCP_CONG_RTT_STAMP 0x2
661
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700662struct tcp_congestion_ops {
663 struct list_head list;
Stephen Hemminger164891a2007-04-23 22:26:16 -0700664 unsigned long flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700665
666 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300667 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700668 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300669 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700670
671 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300672 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700673 /* lower bound for congestion window (optional) */
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700674 u32 (*min_cwnd)(const struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700675 /* do new cwnd calculation (required) */
Ilpo Järvinenc3a05c62007-12-02 00:47:59 +0200676 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700677 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300678 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700679 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300680 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700681 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300682 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700683 /* hook for packet ack accounting (optional) */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700684 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300685 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300686 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700687
688 char name[TCP_CA_NAME_MAX];
689 struct module *owner;
690};
691
692extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
693extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
694
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300695extern void tcp_init_congestion_control(struct sock *sk);
696extern void tcp_cleanup_congestion_control(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700697extern int tcp_set_default_congestion_control(const char *name);
698extern void tcp_get_default_congestion_control(char *name);
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800699extern void tcp_get_available_congestion_control(char *buf, size_t len);
Stephen Hemmingerce7bc3b2006-11-09 16:35:15 -0800700extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
701extern int tcp_set_allowed_congestion_control(char *allowed);
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300702extern int tcp_set_congestion_control(struct sock *sk, const char *name);
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800703extern void tcp_slow_start(struct tcp_sock *tp);
Ilpo Järvinen758ce5c2009-02-28 04:44:37 +0000704extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700705
Stephen Hemminger5f8ef482005-06-23 20:37:36 -0700706extern struct tcp_congestion_ops tcp_init_congestion_ops;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300707extern u32 tcp_reno_ssthresh(struct sock *sk);
Ilpo Järvinenc3a05c62007-12-02 00:47:59 +0200708extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700709extern u32 tcp_reno_min_cwnd(const struct sock *sk);
David S. Millera8acfba2005-06-23 23:45:02 -0700710extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700711
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300712static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700713{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300714 struct inet_connection_sock *icsk = inet_csk(sk);
715
716 if (icsk->icsk_ca_ops->set_state)
717 icsk->icsk_ca_ops->set_state(sk, ca_state);
718 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700719}
720
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300721static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700722{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300723 const struct inet_connection_sock *icsk = inet_csk(sk);
724
725 if (icsk->icsk_ca_ops->cwnd_event)
726 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700727}
728
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300729/* These functions determine how the current flow behaves in respect of SACK
730 * handling. SACK is negotiated with the peer, and therefore it can vary
731 * between different flows.
732 *
733 * tcp_is_sack - SACK enabled
734 * tcp_is_reno - No SACK
735 * tcp_is_fack - FACK enabled, implies SACK enabled
736 */
737static inline int tcp_is_sack(const struct tcp_sock *tp)
738{
739 return tp->rx_opt.sack_ok;
740}
741
742static inline int tcp_is_reno(const struct tcp_sock *tp)
743{
744 return !tcp_is_sack(tp);
745}
746
747static inline int tcp_is_fack(const struct tcp_sock *tp)
748{
749 return tp->rx_opt.sack_ok & 2;
750}
751
752static inline void tcp_enable_fack(struct tcp_sock *tp)
753{
754 tp->rx_opt.sack_ok |= 2;
755}
756
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300757static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
758{
759 return tp->sacked_out + tp->lost_out;
760}
761
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762/* This determines how many packets are "in the network" to the best
763 * of our knowledge. In many cases it is conservative, but where
764 * detailed information is available from the receiver (via SACK
765 * blocks etc.) we can make more aggressive calculations.
766 *
767 * Use this for decisions involving congestion control, use just
768 * tp->packets_out to determine if the send queue is empty or not.
769 *
770 * Read this equation as:
771 *
772 * "Packets sent once on transmission queue" MINUS
773 * "Packets left network, but not honestly ACKed yet" PLUS
774 * "Packets fast retransmitted"
775 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800776static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300778 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779}
780
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -0700781#define TCP_INFINITE_SSTHRESH 0x7fffffff
782
783static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
784{
785 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
786}
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
789 * The exception is rate halving phase, when cwnd is decreasing towards
790 * ssthresh.
791 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300792static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300794 const struct tcp_sock *tp = tcp_sk(sk);
795 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return tp->snd_ssthresh;
797 else
798 return max(tp->snd_ssthresh,
799 ((tp->snd_cwnd >> 1) +
800 (tp->snd_cwnd >> 2)));
801}
802
Ilpo Järvinenb9c45952007-07-27 16:36:17 +0300803/* Use define here intentionally to get WARN_ON location shown at the caller */
804#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
Ilpo Järvinen3cfe3ba2007-02-27 10:09:49 -0800806extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
808
809/* Slow start with delack produces 3 packets of burst, so that
John Heffnerdd9e0dd2008-04-15 15:26:39 -0700810 * it is safe "de facto". This will be the default - same as
811 * the default reordering threshold - but if reordering increases,
812 * we must be able to allow cwnd to burst at least this much in order
813 * to not pull it back when holes are filled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 */
815static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
816{
John Heffnerdd9e0dd2008-04-15 15:26:39 -0700817 return tp->reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
819
Ilpo Järvinen90840de2007-12-31 04:48:41 -0800820/* Returns end sequence number of the receiver's advertised window */
821static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
822{
823 return tp->snd_una + tp->snd_wnd;
824}
Ilpo Järvinencea14e02008-01-12 03:19:12 -0800825extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800826
Chuck Leverc1bd24b2007-10-23 21:08:54 -0700827static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800828 const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
830 if (skb->len < mss)
831 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
832}
833
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700834static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700836 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700837 const struct inet_connection_sock *icsk = inet_csk(sk);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700838
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700839 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700840 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
841 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
Hantzis Fotisee7537b2009-03-02 22:42:02 -0800844static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845{
846 tp->snd_wl1 = seq;
847}
848
Hantzis Fotisee7537b2009-03-02 22:42:02 -0800849static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850{
851 tp->snd_wl1 = seq;
852}
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854/*
855 * Calculate(/check) TCP checksum
856 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -0800857static inline __sum16 tcp_v4_check(int len, __be32 saddr,
858 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
861}
862
Al Virob51655b2006-11-14 21:40:42 -0800863static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Herbert Xufb286bb2005-11-10 13:01:24 -0800865 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866}
867
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800868static inline int tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
Herbert Xu60476372007-04-09 11:59:39 -0700870 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 __tcp_checksum_complete(skb);
872}
873
874/* Prequeue for VJ style copy to user, combined with checksumming. */
875
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800876static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
878 tp->ucopy.task = NULL;
879 tp->ucopy.len = 0;
880 tp->ucopy.memory = 0;
881 skb_queue_head_init(&tp->ucopy.prequeue);
Chris Leech97fc2f02006-05-23 17:55:33 -0700882#ifdef CONFIG_NET_DMA
883 tp->ucopy.dma_chan = NULL;
884 tp->ucopy.wakeup = 0;
885 tp->ucopy.pinned_list = NULL;
886 tp->ucopy.dma_cookie = 0;
887#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}
889
890/* Packet is added to VJ-style prequeue for processing in process
891 * context, if a reader task is waiting. Apparently, this exciting
892 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
893 * failed somewhere. Latency? Burstiness? Well, at least now we will
894 * see, why it failed. 8)8) --ANK
895 *
896 * NOTE: is this not too big to inline?
897 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800898static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899{
900 struct tcp_sock *tp = tcp_sk(sk);
901
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000902 if (sysctl_tcp_low_latency || !tp->ucopy.task)
903 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000905 __skb_queue_tail(&tp->ucopy.prequeue, skb);
906 tp->ucopy.memory += skb->truesize;
907 if (tp->ucopy.memory > sk->sk_rcvbuf) {
908 struct sk_buff *skb1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000910 BUG_ON(sock_owned_by_user(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000912 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
913 sk_backlog_rcv(sk, skb1);
914 NET_INC_STATS_BH(sock_net(sk),
915 LINUX_MIB_TCPPREQUEUEDROPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000917
918 tp->ucopy.memory = 0;
919 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
Eric Dumazetaa395142010-04-20 13:03:51 +0000920 wake_up_interruptible_sync_poll(sk_sleep(sk),
Eric Dumazet7aedec22009-05-07 07:20:39 +0000921 POLLIN | POLLRDNORM | POLLRDBAND);
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000922 if (!inet_csk_ack_scheduled(sk))
923 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
David S. Miller22f6dac2009-05-08 02:48:30 -0700924 (3 * tcp_rto_min(sk)) / 4,
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000925 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 }
Eric Dumazetf5f8d862009-05-07 07:08:38 +0000927 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
930
931#undef STATE_TRACE
932
933#ifdef STATE_TRACE
934static const char *statename[]={
935 "Unused","Established","Syn Sent","Syn Recv",
936 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
937 "Close Wait","Last ACK","Listen","Closing"
938};
939#endif
Ilpo Järvinen490d5042008-01-12 03:17:20 -0800940extern void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Andi Kleen4ac02ba2007-04-20 17:11:46 -0700942extern void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800944static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945{
946 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 rx_opt->num_sacks = 0;
948}
949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950/* Determine a window scaling and initial window to offer. */
951extern void tcp_select_initial_window(int __space, __u32 mss,
952 __u32 *rcv_wnd, __u32 *window_clamp,
laurent chavey31d12922009-12-15 11:15:28 +0000953 int wscale_ok, __u8 *rcv_wscale,
954 __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
956static inline int tcp_win_from_space(int space)
957{
958 return sysctl_tcp_adv_win_scale<=0 ?
959 (space>>(-sysctl_tcp_adv_win_scale)) :
960 space - (space>>sysctl_tcp_adv_win_scale);
961}
962
963/* Note: caller must be prepared to deal with negative returns */
964static inline int tcp_space(const struct sock *sk)
965{
966 return tcp_win_from_space(sk->sk_rcvbuf -
967 atomic_read(&sk->sk_rmem_alloc));
968}
969
970static inline int tcp_full_space(const struct sock *sk)
971{
972 return tcp_win_from_space(sk->sk_rcvbuf);
973}
974
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800975static inline void tcp_openreq_init(struct request_sock *req,
976 struct tcp_options_received *rx_opt,
977 struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700979 struct inet_request_sock *ireq = inet_rsk(req);
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Florian Westphal4dfc2812008-04-10 03:12:40 -0700982 req->cookie_ts = 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700983 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 req->mss = rx_opt->mss_clamp;
985 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700986 ireq->tstamp_ok = rx_opt->tstamp_ok;
987 ireq->sack_ok = rx_opt->sack_ok;
988 ireq->snd_wscale = rx_opt->snd_wscale;
989 ireq->wscale_ok = rx_opt->wscale_ok;
990 ireq->acked = 0;
991 ireq->ecn_ok = 0;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700992 ireq->rmt_port = tcp_hdr(skb)->source;
KOVACS Krisztiana3116ac52008-10-01 07:46:49 -0700993 ireq->loc_port = tcp_hdr(skb)->dest;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
995
Pavel Emelyanov5c52ba12008-07-16 20:28:10 -0700996extern void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998static inline int keepalive_intvl_when(const struct tcp_sock *tp)
999{
1000 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1001}
1002
1003static inline int keepalive_time_when(const struct tcp_sock *tp)
1004{
1005 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1006}
1007
Eric Dumazetdf19a622009-08-28 23:48:54 -07001008static inline int keepalive_probes(const struct tcp_sock *tp)
1009{
1010 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1011}
1012
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001013static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1014{
1015 const struct inet_connection_sock *icsk = &tp->inet_conn;
1016
1017 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1018 tcp_time_stamp - tp->rcv_tstamp);
1019}
1020
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001021static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001023 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1024 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001026 if (fin_timeout < (rto << 2) - (rto >> 1))
1027 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 return fin_timeout;
1030}
1031
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001032static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1033 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001035 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1036 return 1;
1037 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1038 return 1;
1039
1040 return 0;
1041}
1042
1043static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1044 int rst)
1045{
1046 if (tcp_paws_check(rx_opt, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 return 0;
1048
1049 /* RST segments are not recommended to carry timestamp,
1050 and, if they do, it is recommended to ignore PAWS because
1051 "their cleanup function should take precedence over timestamps."
1052 Certainly, it is mistake. It is necessary to understand the reasons
1053 of this constraint to relax it: if peer reboots, clock may go
1054 out-of-sync and half-open connections will not be reset.
1055 Actually, the problem would be not existing if all
1056 the implementations followed draft about maintaining clock
1057 via reboots. Linux-2.2 DOES NOT!
1058
1059 However, we can relax time bounds for RST segments to MSL.
1060 */
James Morris9d729f72007-03-04 16:12:44 -08001061 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 return 0;
1063 return 1;
1064}
1065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066#define TCP_CHECK_TIMER(sk) do { } while (0)
1067
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001068static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069{
1070 /* See RFC 2012 */
Pavel Emelyanovcf1100a2008-07-16 20:27:38 -07001071 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1072 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1073 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1074 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001077/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001078static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001079{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001080 tp->lost_skb_hint = NULL;
1081 tp->scoreboard_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001082}
1083
1084static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1085{
1086 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001087 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001088}
1089
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001090/* MD5 Signature */
1091struct crypto_hash;
1092
1093/* - key database */
1094struct tcp_md5sig_key {
1095 u8 *key;
1096 u8 keylen;
1097};
1098
1099struct tcp4_md5sig_key {
David S. Millerf8ab18d2007-09-28 15:18:35 -07001100 struct tcp_md5sig_key base;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001101 __be32 addr;
1102};
1103
1104struct tcp6_md5sig_key {
David S. Millerf8ab18d2007-09-28 15:18:35 -07001105 struct tcp_md5sig_key base;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001106#if 0
1107 u32 scope_id; /* XXX */
1108#endif
1109 struct in6_addr addr;
1110};
1111
1112/* - sock block */
1113struct tcp_md5sig_info {
1114 struct tcp4_md5sig_key *keys4;
1115#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1116 struct tcp6_md5sig_key *keys6;
1117 u32 entries6;
1118 u32 alloced6;
1119#endif
1120 u32 entries4;
1121 u32 alloced4;
1122};
1123
1124/* - pseudo header */
1125struct tcp4_pseudohdr {
1126 __be32 saddr;
1127 __be32 daddr;
1128 __u8 pad;
1129 __u8 protocol;
1130 __be16 len;
1131};
1132
1133struct tcp6_pseudohdr {
1134 struct in6_addr saddr;
1135 struct in6_addr daddr;
1136 __be32 len;
1137 __be32 protocol; /* including padding */
1138};
1139
1140union tcp_md5sum_block {
1141 struct tcp4_pseudohdr ip4;
1142#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1143 struct tcp6_pseudohdr ip6;
1144#endif
1145};
1146
1147/* - pool: digest algorithm, hash description and scratch buffer */
1148struct tcp_md5sig_pool {
1149 struct hash_desc md5_desc;
1150 union tcp_md5sum_block md5_blk;
1151};
1152
1153#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1154
1155/* - functions */
Changli Gao53d31762010-07-10 20:41:06 +00001156extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1157 struct sock *sk, struct request_sock *req,
1158 struct sk_buff *skb);
1159extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1160 struct sock *addr_sk);
1161extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1162 u8 newkeylen);
1163extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001164
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001165#ifdef CONFIG_TCP_MD5SIG
1166#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1167 &(struct tcp_md5sig_key) { \
1168 .key = (twsk)->tw_md5_key, \
1169 .keylen = (twsk)->tw_md5_keylen, \
1170 } : NULL)
1171#else
1172#define tcp_twsk_md5_key(twsk) NULL
1173#endif
1174
Tejun Heo7d720c32010-02-16 15:20:26 +00001175extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
Changli Gao53d31762010-07-10 20:41:06 +00001176extern void tcp_free_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001177
Eric Dumazet35790c02010-05-16 00:34:04 -07001178extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Changli Gao53d31762010-07-10 20:41:06 +00001179extern void tcp_put_md5sig_pool(void);
Eric Dumazet35790c02010-05-16 00:34:04 -07001180
Adam Langley49a72df2008-07-19 00:01:42 -07001181extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1182extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1183 unsigned header_len);
1184extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1185 struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001186
David S. Millerfe067e82007-03-07 12:12:44 -08001187/* write queue abstraction */
1188static inline void tcp_write_queue_purge(struct sock *sk)
1189{
1190 struct sk_buff *skb;
1191
1192 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001193 sk_wmem_free_skb(sk, skb);
1194 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001195 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001196}
1197
1198static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1199{
David S. Millercd07a8e2008-09-23 00:50:13 -07001200 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001201}
1202
1203static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1204{
David S. Millercd07a8e2008-09-23 00:50:13 -07001205 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001206}
1207
1208static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1209{
David S. Millercd07a8e2008-09-23 00:50:13 -07001210 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001211}
1212
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001213static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1214{
1215 return skb_queue_prev(&sk->sk_write_queue, skb);
1216}
1217
David S. Millerfe067e82007-03-07 12:12:44 -08001218#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001219 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001220
1221#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001222 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001223
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001224#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001225 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001226
David S. Millerfe067e82007-03-07 12:12:44 -08001227static inline struct sk_buff *tcp_send_head(struct sock *sk)
1228{
1229 return sk->sk_send_head;
1230}
1231
David S. Millercd07a8e2008-09-23 00:50:13 -07001232static inline bool tcp_skb_is_last(const struct sock *sk,
1233 const struct sk_buff *skb)
1234{
1235 return skb_queue_is_last(&sk->sk_write_queue, skb);
1236}
1237
David S. Millerfe067e82007-03-07 12:12:44 -08001238static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1239{
David S. Millercd07a8e2008-09-23 00:50:13 -07001240 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001241 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001242 else
1243 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001244}
1245
1246static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1247{
1248 if (sk->sk_send_head == skb_unlinked)
1249 sk->sk_send_head = NULL;
1250}
1251
1252static inline void tcp_init_send_head(struct sock *sk)
1253{
1254 sk->sk_send_head = NULL;
1255}
1256
1257static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1258{
1259 __skb_queue_tail(&sk->sk_write_queue, skb);
1260}
1261
1262static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1263{
1264 __tcp_add_write_queue_tail(sk, skb);
1265
1266 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001267 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001268 sk->sk_send_head = skb;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001269
1270 if (tcp_sk(sk)->highest_sack == NULL)
1271 tcp_sk(sk)->highest_sack = skb;
1272 }
David S. Millerfe067e82007-03-07 12:12:44 -08001273}
1274
1275static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1276{
1277 __skb_queue_head(&sk->sk_write_queue, skb);
1278}
1279
1280/* Insert buff after skb on the write queue of sk. */
1281static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1282 struct sk_buff *buff,
1283 struct sock *sk)
1284{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001285 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001286}
1287
David S. Miller43f59c82008-09-21 21:28:51 -07001288/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001289static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1290 struct sk_buff *skb,
1291 struct sock *sk)
1292{
David S. Miller43f59c82008-09-21 21:28:51 -07001293 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001294
1295 if (sk->sk_send_head == skb)
1296 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001297}
1298
1299static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1300{
1301 __skb_unlink(skb, &sk->sk_write_queue);
1302}
1303
David S. Millerfe067e82007-03-07 12:12:44 -08001304static inline int tcp_write_queue_empty(struct sock *sk)
1305{
1306 return skb_queue_empty(&sk->sk_write_queue);
1307}
1308
Krishna Kumar12d50c42009-12-08 22:26:13 +00001309static inline void tcp_push_pending_frames(struct sock *sk)
1310{
1311 if (tcp_send_head(sk)) {
1312 struct tcp_sock *tp = tcp_sk(sk);
1313
1314 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1315 }
1316}
1317
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001318/* Start sequence of the highest skb with SACKed bit, valid only if
1319 * sacked > 0 or when the caller has ensured validity by itself.
1320 */
1321static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1322{
1323 if (!tp->sacked_out)
1324 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001325
1326 if (tp->highest_sack == NULL)
1327 return tp->snd_nxt;
1328
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001329 return TCP_SKB_CB(tp->highest_sack)->seq;
1330}
1331
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001332static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1333{
1334 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1335 tcp_write_queue_next(sk, skb);
1336}
1337
1338static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1339{
1340 return tcp_sk(sk)->highest_sack;
1341}
1342
1343static inline void tcp_highest_sack_reset(struct sock *sk)
1344{
1345 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1346}
1347
1348/* Called when old skb is about to be deleted (to be combined with new skb) */
1349static inline void tcp_highest_sack_combine(struct sock *sk,
1350 struct sk_buff *old,
1351 struct sk_buff *new)
1352{
1353 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1354 tcp_sk(sk)->highest_sack = new;
1355}
1356
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001357/* Determines whether this is a thin stream (which may suffer from
1358 * increased latency). Used to trigger latency-reducing mechanisms.
1359 */
1360static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1361{
1362 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1363}
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365/* /proc */
1366enum tcp_seq_states {
1367 TCP_SEQ_STATE_LISTENING,
1368 TCP_SEQ_STATE_OPENREQ,
1369 TCP_SEQ_STATE_ESTABLISHED,
1370 TCP_SEQ_STATE_TIME_WAIT,
1371};
1372
1373struct tcp_seq_afinfo {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 char *name;
1375 sa_family_t family;
Denis V. Lunev68fcadd2008-04-13 22:13:30 -07001376 struct file_operations seq_fops;
Denis V. Lunev9427c4b2008-04-13 22:12:13 -07001377 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378};
1379
1380struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001381 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 sa_family_t family;
1383 enum tcp_seq_states state;
1384 struct sock *syn_wait_sk;
Tom Herberta8b690f2010-06-07 00:43:42 -07001385 int bucket, offset, sbucket, num, uid;
1386 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387};
1388
Daniel Lezcano6f8b13b2008-03-21 04:14:45 -07001389extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1390extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001392extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001393extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001394
Brian Haley7d06b2e2008-06-14 17:04:49 -07001395extern void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001396
Herbert Xua430a432006-07-08 13:34:56 -07001397extern int tcp_v4_gso_send_check(struct sk_buff *skb);
Herbert Xu576a30e2006-06-27 13:22:38 -07001398extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
Herbert Xubf296b12008-12-15 23:43:36 -08001399extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1400 struct sk_buff *skb);
1401extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1402 struct sk_buff *skb);
1403extern int tcp_gro_complete(struct sk_buff *skb);
1404extern int tcp4_gro_complete(struct sk_buff *skb);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001405
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001406#ifdef CONFIG_PROC_FS
Changli Gao53d31762010-07-10 20:41:06 +00001407extern int tcp4_proc_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001408extern void tcp4_proc_exit(void);
1409#endif
1410
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001411/* TCP af-specific functions */
1412struct tcp_sock_af_ops {
1413#ifdef CONFIG_TCP_MD5SIG
1414 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1415 struct sock *addr_sk);
1416 int (*calc_md5_hash) (char *location,
1417 struct tcp_md5sig_key *md5,
1418 struct sock *sk,
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001419 struct request_sock *req,
Adam Langley49a72df2008-07-19 00:01:42 -07001420 struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001421 int (*md5_add) (struct sock *sk,
1422 struct sock *addr_sk,
1423 u8 *newkey,
1424 u8 len);
1425 int (*md5_parse) (struct sock *sk,
1426 char __user *optval,
1427 int optlen);
1428#endif
1429};
1430
1431struct tcp_request_sock_ops {
1432#ifdef CONFIG_TCP_MD5SIG
1433 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1434 struct request_sock *req);
John Dykstrae3afe7b2009-07-16 05:04:51 +00001435 int (*calc_md5_hash) (char *location,
1436 struct tcp_md5sig_key *md5,
1437 struct sock *sk,
1438 struct request_sock *req,
1439 struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001440#endif
1441};
1442
William Allen Simpsonda5c78c2009-12-02 18:12:09 +00001443/* Using SHA1 for now, define some constants.
1444 */
1445#define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1446#define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1447#define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1448
1449extern int tcp_cookie_generator(u32 *bakery);
1450
William Allen Simpson435cf552009-12-02 18:17:05 +00001451/**
1452 * struct tcp_cookie_values - each socket needs extra space for the
1453 * cookies, together with (optional) space for any SYN data.
1454 *
1455 * A tcp_sock contains a pointer to the current value, and this is
1456 * cloned to the tcp_timewait_sock.
1457 *
1458 * @cookie_pair: variable data from the option exchange.
1459 *
1460 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1461 * indicates default (sysctl_tcp_cookie_size).
1462 * After cookie sent, remembers size of cookie.
1463 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1464 *
1465 * @s_data_desired: user specified tcpct_s_data_desired. When the
1466 * constant payload is specified (@s_data_constant),
1467 * holds its length instead.
1468 * Range 0 to TCP_MSS_DESIRED.
1469 *
1470 * @s_data_payload: constant data that is to be included in the
1471 * payload of SYN or SYNACK segments when the
1472 * cookie option is present.
1473 */
1474struct tcp_cookie_values {
1475 struct kref kref;
1476 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1477 u8 cookie_pair_size;
1478 u8 cookie_desired;
1479 u16 s_data_desired:11,
1480 s_data_constant:1,
1481 s_data_in:1,
1482 s_data_out:1,
1483 s_data_unused:2;
1484 u8 s_data_payload[0];
1485};
1486
1487static inline void tcp_cookie_values_release(struct kref *kref)
1488{
1489 kfree(container_of(kref, struct tcp_cookie_values, kref));
1490}
1491
1492/* The length of constant payload data. Note that s_data_desired is
1493 * overloaded, depending on s_data_constant: either the length of constant
1494 * data (returned here) or the limit on variable data.
1495 */
1496static inline int tcp_s_data_size(const struct tcp_sock *tp)
1497{
1498 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1499 ? tp->cookie_values->s_data_desired
1500 : 0;
1501}
1502
1503/**
1504 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1505 *
1506 * As tcp_request_sock has already been extended in other places, the
1507 * only remaining method is to pass stack values along as function
1508 * parameters. These parameters are not needed after sending SYNACK.
1509 *
1510 * @cookie_bakery: cryptographic secret and message workspace.
1511 *
1512 * @cookie_plus: bytes in authenticator/cookie option, copied from
1513 * struct tcp_options_received (above).
1514 */
1515struct tcp_extend_values {
1516 struct request_values rv;
1517 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1518 u8 cookie_plus:6,
1519 cookie_out_never:1,
1520 cookie_in_always:1;
1521};
1522
1523static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1524{
1525 return (struct tcp_extend_values *)rvp;
1526}
1527
Denis V. Lunev9b0f9762008-02-29 11:13:15 -08001528extern void tcp_v4_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001529extern void tcp_init(void);
1530
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531#endif /* _TCP_H */