blob: 7292c3c575bc9eddf2736199776fa6df0c406ab5 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#define FASTRETRANS_DEBUG 1
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/list.h>
24#include <linux/tcp.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050025#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#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>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080030#include <linux/crypto.h>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080031#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000032#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080033#include <linux/ktime.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>
Glauber Costa180d8cd2011-12-11 21:47:02 +000048#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070050extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Eric Dumazetdd24c002008-11-25 21:17:14 -080052extern struct percpu_counter tcp_orphan_count;
Joe Perches5c9f3022013-09-23 11:33:32 -070053void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070056#define MAX_TCP_OPTION_SPACE 40
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090058/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090060 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 */
62#define MAX_TCP_WINDOW 32767U
63
64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65#define TCP_MIN_MSS 88U
66
John Heffner5d424d52006-03-20 17:53:41 -080067/* The least MTU to use for probing */
Fan Dudcd8fb82015-03-06 11:18:22 +080068#define TCP_BASE_MSS 1024
John Heffner5d424d52006-03-20 17:53:41 -080069
Fan Du05cbc0d2015-03-06 11:18:24 +080070/* probing interval, default to 10 minutes as per RFC4821 */
71#define TCP_PROBE_INTERVAL 600
72
Fan Du6b58e0a2015-03-06 11:18:23 +080073/* Specify interval when tcp mtu probing will stop */
74#define TCP_PROBE_THRESHOLD 8
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* After receiving this amount of duplicate ACKs fast retransmit starts. */
77#define TCP_FASTRETRANS_THRESH 3
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* Maximal number of ACKs sent quickly to accelerate slow-start. */
80#define TCP_MAX_QUICKACKS 16U
81
82/* urg_data states */
83#define TCP_URG_VALID 0x0100
84#define TCP_URG_NOTYET 0x0200
85#define TCP_URG_READ 0x0400
86
87#define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
92 */
93
94#define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
99 */
100
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000101#define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
108 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000110#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
114 * initial RTO.
115 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
123 * TIME-WAIT timer.
124 */
125
126#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
127#if HZ >= 100
128#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129#define TCP_ATO_MIN ((unsigned)(HZ/25))
130#else
131#define TCP_DELACK_MIN 4U
132#define TCP_ATO_MIN 4U
133#endif
134#define TCP_RTO_MAX ((unsigned)(120*HZ))
135#define TCP_RTO_MIN ((unsigned)(HZ/5))
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000136#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000137#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
138 * used as a fallback RTO for the
139 * initial data transmission if no
140 * valid RTT sample has been acquired,
141 * most likely due to retrans in 3WHS.
142 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
144#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
145 * for local resources.
146 */
147
148#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150#define TCP_KEEPALIVE_INTVL (75*HZ)
151
152#define MAX_TCP_KEEPIDLE 32767
153#define MAX_TCP_KEEPINTVL 32767
154#define MAX_TCP_KEEPCNT 127
155#define MAX_TCP_SYNCNT 127
156
157#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
165 */
166#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/*
171 * TCP option
172 */
Kenjiro Nakayama105970f2014-10-20 18:15:50 +0900173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174#define TCPOPT_NOP 1 /* Padding */
175#define TCPOPT_EOL 0 /* End of options */
176#define TCPOPT_MSS 2 /* Segment size negotiating */
177#define TCPOPT_WINDOW 3 /* Window scaling */
178#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179#define TCPOPT_SACK 5 /* SACK Block */
180#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800181#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Daniel Lee7f9b8382015-04-06 14:37:26 -0700182#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000183#define TCPOPT_EXP 254 /* Experimental */
184/* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
186 */
187#define TCPOPT_FASTOPEN_MAGIC 0xF989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189/*
190 * TCP option lengths
191 */
192
193#define TCPOLEN_MSS 4
194#define TCPOLEN_WINDOW 3
195#define TCPOLEN_SACK_PERM 2
196#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800197#define TCPOLEN_MD5SIG 18
Daniel Lee7f9b8382015-04-06 14:37:26 -0700198#define TCPOLEN_FASTOPEN_BASE 2
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000199#define TCPOLEN_EXP_FASTOPEN_BASE 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201/* But this is what stacks really send out. */
202#define TCPOLEN_TSTAMP_ALIGNED 12
203#define TCPOLEN_WSCALE_ALIGNED 4
204#define TCPOLEN_SACKPERM_ALIGNED 4
205#define TCPOLEN_SACK_BASE 2
206#define TCPOLEN_SACK_BASE_ALIGNED 4
207#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800208#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700209#define TCPOLEN_MSS_ALIGNED 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211/* Flags in tp->nonagle */
212#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
213#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800214#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000216/* TCP thin-stream limits */
217#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
218
David S. Miller7eb38522011-02-05 18:13:45 -0800219/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
David S. Miller442b9632011-02-02 17:05:11 -0800220#define TCP_INIT_CWND 10
221
Yuchung Chengcf60af02012-07-19 06:43:09 +0000222/* Bit Flags for sysctl_tcp_fastopen */
223#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000224#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000225#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af02012-07-19 06:43:09 +0000226
Jerry Chu10467162012-08-31 12:29:11 +0000227/* Accept SYN data w/o any cookie option */
228#define TFO_SERVER_COOKIE_NOT_REQD 0x200
229
230/* Force enable TFO on all listeners, i.e., not requiring the
231 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
232 */
233#define TFO_SERVER_WO_SOCKOPT1 0x400
234#define TFO_SERVER_WO_SOCKOPT2 0x800
Jerry Chu10467162012-08-31 12:29:11 +0000235
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700236extern struct inet_timewait_death_row tcp_death_row;
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239extern int sysctl_tcp_timestamps;
240extern int sysctl_tcp_window_scaling;
241extern int sysctl_tcp_sack;
242extern int sysctl_tcp_fin_timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243extern int sysctl_tcp_keepalive_time;
244extern int sysctl_tcp_keepalive_probes;
245extern int sysctl_tcp_keepalive_intvl;
246extern int sysctl_tcp_syn_retries;
247extern int sysctl_tcp_synack_retries;
248extern int sysctl_tcp_retries1;
249extern int sysctl_tcp_retries2;
250extern int sysctl_tcp_orphan_retries;
251extern int sysctl_tcp_syncookies;
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000252extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253extern int sysctl_tcp_retrans_collapse;
254extern int sysctl_tcp_stdurg;
255extern int sysctl_tcp_rfc1337;
256extern int sysctl_tcp_abort_on_overflow;
257extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258extern int sysctl_tcp_fack;
259extern int sysctl_tcp_reordering;
Eric Dumazetdca145f2014-10-27 21:45:24 -0700260extern int sysctl_tcp_max_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700262extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263extern int sysctl_tcp_wmem[3];
264extern int sysctl_tcp_rmem[3];
265extern int sysctl_tcp_app_win;
266extern int sysctl_tcp_adv_win_scale;
267extern int sysctl_tcp_tw_reuse;
268extern int sysctl_tcp_frto;
269extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271extern int sysctl_tcp_moderate_rcvbuf;
272extern int sysctl_tcp_tso_win_divisor;
Rick Jones15d99e02006-03-20 22:40:29 -0800273extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700274extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000275extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000276extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000277extern int sysctl_tcp_early_retrans;
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000278extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200279extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -0700280extern unsigned int sysctl_tcp_notsent_lowat;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700281extern int sysctl_tcp_min_tso_segs;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800282extern int sysctl_tcp_autocorking;
Neal Cardwell032ee422015-02-06 16:04:38 -0500283extern int sysctl_tcp_invalid_ratelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Eric Dumazet8d987e52010-11-09 23:24:26 +0000285extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800286extern struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287extern int tcp_memory_pressure;
288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * The next routines deal with comparing 32 bit unsigned ints
291 * and worry about wraparound (automatic with unsigned arithmetic).
292 */
293
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000294static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800296 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800298#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000301static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 return seq3 - seq2 >= seq1 - seq2;
304}
305
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800306static inline bool tcp_out_of_memory(struct sock *sk)
307{
308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
310 return true;
311 return false;
312}
313
David S. Millerad1af0f2010-08-25 02:27:49 -0700314static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700315{
David S. Millerad1af0f2010-08-25 02:27:49 -0700316 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
317 int orphans = percpu_counter_read_positive(ocp);
318
319 if (orphans << shift > sysctl_tcp_max_orphans) {
320 orphans = percpu_counter_sum_positive(ocp);
321 if (orphans << shift > sysctl_tcp_max_orphans)
322 return true;
323 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700324 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700325}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Joe Perches5c9f3022013-09-23 11:33:32 -0700327bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800328
Florian Westphala0f82f62009-04-19 09:43:48 +0000329/* syncookies: remember time of last synqueue overflow */
330static inline void tcp_synq_overflow(struct sock *sk)
331{
332 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
333}
334
335/* syncookies: no recent synqueue overflow on this listening socket? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000336static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
Florian Westphala0f82f62009-04-19 09:43:48 +0000337{
338 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
Jerry Chu9ad7c042011-06-08 11:08:38 +0000339 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
Florian Westphala0f82f62009-04-19 09:43:48 +0000340}
341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342extern struct proto tcp_prot;
343
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700344#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
345#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
346#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
347#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 +0000348#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
Joe Perches5c9f3022013-09-23 11:33:32 -0700350void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000351
Joe Perches5c9f3022013-09-23 11:33:32 -0700352void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
Joe Perches5c9f3022013-09-23 11:33:32 -0700354void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
Joe Perches5c9f3022013-09-23 11:33:32 -0700356void tcp_v4_early_demux(struct sk_buff *skb);
357int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
Joe Perches5c9f3022013-09-23 11:33:32 -0700359int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800360int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700361int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
362 int flags);
363void tcp_release_cb(struct sock *sk);
364void tcp_wfree(struct sk_buff *skb);
365void tcp_write_timer_handler(struct sock *sk);
366void tcp_delack_timer_handler(struct sock *sk);
367int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
368int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
369 const struct tcphdr *th, unsigned int len);
370void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
371 const struct tcphdr *th, unsigned int len);
372void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700373int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
374void tcp_twsk_destructor(struct sock *sk);
375ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
376 struct pipe_inode_info *pipe, size_t len,
377 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800378
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700379static inline void tcp_dec_quickack_mode(struct sock *sk,
380 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700382 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700383
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700384 if (icsk->icsk_ack.quick) {
385 if (pkts >= icsk->icsk_ack.quick) {
386 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700387 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700388 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700389 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700390 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 }
392}
393
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700394#define TCP_ECN_OK 1
395#define TCP_ECN_QUEUE_CWR 2
396#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000397#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700398
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000399enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 TCP_TW_SUCCESS = 0,
401 TCP_TW_RST = 1,
402 TCP_TW_ACK = 2,
403 TCP_TW_SYN = 3
404};
405
406
Joe Perches5c9f3022013-09-23 11:33:32 -0700407enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
408 struct sk_buff *skb,
409 const struct tcphdr *th);
410struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazet52452c52015-03-19 19:04:19 -0700411 struct request_sock *req, bool fastopen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700412int tcp_child_process(struct sock *parent, struct sock *child,
413 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400414void tcp_enter_loss(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700415void tcp_clear_retrans(struct tcp_sock *tp);
416void tcp_update_metrics(struct sock *sk);
417void tcp_init_metrics(struct sock *sk);
418void tcp_metrics_init(void);
419bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200420 bool paws_check, bool timestamps);
Joe Perches5c9f3022013-09-23 11:33:32 -0700421bool tcp_remember_stamp(struct sock *sk);
422bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
423void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
424void tcp_disable_fack(struct tcp_sock *tp);
425void tcp_close(struct sock *sk, long timeout);
426void tcp_init_sock(struct sock *sk);
427unsigned int tcp_poll(struct file *file, struct socket *sock,
428 struct poll_table_struct *wait);
429int tcp_getsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, int __user *optlen);
431int tcp_setsockopt(struct sock *sk, int level, int optname,
432 char __user *optval, unsigned int optlen);
433int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000434 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700435int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000436 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700437void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700438void tcp_syn_ack_timeout(const struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800439int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
440 int flags, int *addr_len);
Joe Perches5c9f3022013-09-23 11:33:32 -0700441void tcp_parse_options(const struct sk_buff *skb,
442 struct tcp_options_received *opt_rx,
443 int estab, struct tcp_fastopen_cookie *foc);
444const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446/*
447 * TCP v4 functions exported for the inet6 API
448 */
449
Joe Perches5c9f3022013-09-23 11:33:32 -0700450void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400451void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet26e37362015-03-22 10:22:22 -0700452void tcp_req_err(struct sock *sk, u32 seq);
Joe Perches5c9f3022013-09-23 11:33:32 -0700453int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
454struct sock *tcp_create_openreq_child(struct sock *sk,
455 struct request_sock *req,
456 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100457void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700458struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
459 struct request_sock *req,
460 struct dst_entry *dst);
461int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
462int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
463int tcp_connect(struct sock *sk);
464struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
465 struct request_sock *req,
466 struct tcp_fastopen_cookie *foc);
467int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000469void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000470int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000471void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473/* From syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700474int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
475 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700476struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400477#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200478
Eric Dumazet63262312014-03-19 21:02:21 -0700479/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200480 * This counter is used both as a hash input and partially encoded into
481 * the cookie value. A cookie is only validated further if the delta
482 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700483 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200484 * the counter advances immediately after a cookie is generated).
485 */
486#define MAX_SYNCOOKIE_AGE 2
487
488static inline u32 tcp_cookie_time(void)
489{
Eric Dumazet63262312014-03-19 21:02:21 -0700490 u64 val = get_jiffies_64();
491
492 do_div(val, 60 * HZ);
493 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200494}
495
Joe Perches5c9f3022013-09-23 11:33:32 -0700496u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
497 u16 *mssp);
Octavian Purdila57b47552014-06-25 17:09:50 +0300498__u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
499 __u16 *mss);
Joe Perches5c9f3022013-09-23 11:33:32 -0700500__u32 cookie_init_timestamp(struct request_sock *req);
Florian Westphalf1673382014-11-03 17:35:02 +0100501bool cookie_timestamp_decode(struct tcp_options_received *opt);
502bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100503 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700504
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800505/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700506int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
507 u32 cookie);
508struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100509
Joe Perches5c9f3022013-09-23 11:33:32 -0700510u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
511 const struct tcphdr *th, u16 *mssp);
512__u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
513 __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400514#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515/* tcp_output.c */
516
Joe Perches5c9f3022013-09-23 11:33:32 -0700517void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
518 int nonagle);
519bool tcp_may_send_now(struct sock *sk);
520int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
521int tcp_retransmit_skb(struct sock *, struct sk_buff *);
522void tcp_retransmit_timer(struct sock *sk);
523void tcp_xmit_retransmit_queue(struct sock *);
524void tcp_simple_retransmit(struct sock *);
525int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300526int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Joe Perches5c9f3022013-09-23 11:33:32 -0700528void tcp_send_probe0(struct sock *);
529void tcp_send_partial(struct sock *);
530int tcp_write_wakeup(struct sock *);
531void tcp_send_fin(struct sock *sk);
532void tcp_send_active_reset(struct sock *sk, gfp_t priority);
533int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700534void tcp_push_one(struct sock *, unsigned int mss_now);
535void tcp_send_ack(struct sock *sk);
536void tcp_send_delayed_ack(struct sock *sk);
537void tcp_send_loss_probe(struct sock *sk);
538bool tcp_schedule_loss_probe(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
David S. Millera762a982005-07-05 15:18:51 -0700540/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700541void tcp_resume_early_retransmit(struct sock *sk);
542void tcp_rearm_rto(struct sock *sk);
543void tcp_reset(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700546void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700547static inline void tcp_clear_xmit_timers(struct sock *sk)
548{
549 inet_csk_clear_xmit_timers(sk);
550}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Joe Perches5c9f3022013-09-23 11:33:32 -0700552unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
553unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000554
555/* Bound MSS / TSO packet size with the half of the window */
556static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
557{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700558 int cutoff;
559
560 /* When peer uses tiny windows, there is no use in packetizing
561 * to sub-MSS pieces for the sake of SWS or making sure there
562 * are enough packets in the pipe for fast recovery.
563 *
564 * On the other hand, for extremely large MSS devices, handling
565 * smaller than MSS windows in this way does make sense.
566 */
567 if (tp->max_window >= 512)
568 cutoff = (tp->max_window >> 1);
569 else
570 cutoff = tp->max_window;
571
572 if (cutoff && pktsize > cutoff)
573 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000574 else
575 return pktsize;
576}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300578/* tcp.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700579void tcp_get_info(const struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581/* Read 'sendfile()'-style from a TCP socket */
582typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
583 unsigned int, size_t);
Joe Perches5c9f3022013-09-23 11:33:32 -0700584int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
585 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Joe Perches5c9f3022013-09-23 11:33:32 -0700587void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Joe Perches5c9f3022013-09-23 11:33:32 -0700589int tcp_mtu_to_mss(struct sock *sk, int pmtu);
590int tcp_mss_to_mtu(struct sock *sk, int mss);
591void tcp_mtup_init(struct sock *sk);
592void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800593
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000594static inline void tcp_bound_rto(const struct sock *sk)
595{
596 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
597 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
598}
599
600static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
601{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800602 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000603}
604
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800605static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
607 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
608 ntohl(TCP_FLAG_ACK) |
609 snd_wnd);
610}
611
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800612static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
614 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
615}
616
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700617static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700619 struct tcp_sock *tp = tcp_sk(sk);
620
David S. Millerb03efcf2005-07-08 14:57:23 -0700621 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 tp->rcv_wnd &&
623 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
624 !tp->urg_data)
625 tcp_fast_path_on(tp);
626}
627
Satoru SATOH0c266892009-05-04 11:11:01 -0700628/* Compute the actual rto_min value */
629static inline u32 tcp_rto_min(struct sock *sk)
630{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400631 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700632 u32 rto_min = TCP_RTO_MIN;
633
634 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
635 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
636 return rto_min;
637}
638
Eric Dumazet740b0f12014-02-26 14:02:48 -0800639static inline u32 tcp_rto_min_us(struct sock *sk)
640{
641 return jiffies_to_usecs(tcp_rto_min(sk));
642}
643
Daniel Borkmann81164412015-01-05 23:57:48 +0100644static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
645{
646 return dst_metric_locked(dst, RTAX_CC_ALGO);
647}
648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649/* Compute the actual receive window we are currently advertising.
650 * Rcv_nxt can be after the window if our peer push more data
651 * than the offered window.
652 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800653static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
655 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
656
657 if (win < 0)
658 win = 0;
659 return (u32) win;
660}
661
662/* Choose a new window, without checks for shrinking, and without
663 * scaling applied to the result. The caller does these things
664 * if necessary. This is a "raw" window selection.
665 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700666u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Pavel Emelyanovee995282012-04-19 03:40:39 +0000668void tcp_send_window_probe(struct sock *sk);
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670/* TCP timestamps are only 32-bits, this causes a slight
671 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800672 * of jiffies in the buffer control blocks below. We decided
673 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 * casts with the following macro.
675 */
676#define tcp_time_stamp ((__u32)(jiffies))
677
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700678static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
679{
680 return skb->skb_mstamp.stamp_jiffies;
681}
682
683
Changli Gaoa3433f32010-06-12 14:01:43 +0000684#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
685
686#define TCPHDR_FIN 0x01
687#define TCPHDR_SYN 0x02
688#define TCPHDR_RST 0x04
689#define TCPHDR_PSH 0x08
690#define TCPHDR_ACK 0x10
691#define TCPHDR_URG 0x20
692#define TCPHDR_ECE 0x40
693#define TCPHDR_CWR 0x80
694
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800695/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700696 * TCP per-packet control information to the transmission code.
697 * We also store the host-order sequence numbers in here too.
698 * This is 44 bytes if IPV6 is enabled.
699 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 */
701struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 __u32 seq; /* Starting sequence number */
703 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700704 union {
705 /* Note : tcp_tw_isn is used in input path only
706 * (isn chosen by tcp_timewait_state_process())
707 *
708 * tcp_gso_segs is used in write queue only,
709 * cf tcp_skb_pcount()
710 */
711 __u32 tcp_tw_isn;
712 __u32 tcp_gso_segs;
713 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400714 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 __u8 sacked; /* State flags for SACK/FACK. */
717#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
718#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
719#define TCPCB_LOST 0x04 /* SKB is lost */
720#define TCPCB_TAGBITS 0x07 /* All tag bits */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400721#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400723#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
724 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000726 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
727 /* 1 byte hole */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700729 union {
730 struct inet_skb_parm h4;
731#if IS_ENABLED(CONFIG_IPV6)
732 struct inet6_skb_parm h6;
733#endif
734 } header; /* For incoming frames */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735};
736
737#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
738
Eric Dumazet870c3152014-10-17 09:17:20 -0700739
Eric Dumazet815afe12014-10-18 08:34:37 -0700740#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700741/* This is the variant of inet6_iif() that must be used by TCP,
742 * as TCP moves IP6CB into a different location in skb->cb[]
743 */
744static inline int tcp_v6_iif(const struct sk_buff *skb)
745{
746 return TCP_SKB_CB(skb)->header.h6.iif;
747}
Eric Dumazet815afe12014-10-18 08:34:37 -0700748#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750/* Due to TSO, an SKB can be composed of multiple actual
751 * packets. To keep these tracked properly, we use this.
752 */
753static inline int tcp_skb_pcount(const struct sk_buff *skb)
754{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700755 return TCP_SKB_CB(skb)->tcp_gso_segs;
756}
757
758static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
759{
760 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
761}
762
763static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
764{
765 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
768/* This is valid iff tcp_skb_pcount() > 1. */
769static inline int tcp_skb_mss(const struct sk_buff *skb)
770{
Herbert Xu79671682006-06-22 02:40:14 -0700771 return skb_shinfo(skb)->gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772}
773
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700774/* Events passed to congestion control interface */
775enum tcp_ca_event {
776 CA_EVENT_TX_START, /* first transmit when no packets in flight */
777 CA_EVENT_CWND_RESTART, /* congestion window restart */
778 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700779 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200780 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
781 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
782 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
783 CA_EVENT_NON_DELAYED_ACK,
Florian Westphal7354c8c2014-09-26 22:37:34 +0200784};
785
Florian Westphal98900922014-09-26 22:37:35 +0200786/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200787enum tcp_ca_ack_event_flags {
Florian Westphal98900922014-09-26 22:37:35 +0200788 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
789 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
790 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700791};
792
793/*
794 * Interface for adding new TCP congestion control handlers
795 */
796#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800797#define TCP_CA_MAX 128
798#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
799
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100800#define TCP_CA_UNSPEC 0
801
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200802/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700803#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200804/* Requires ECN/ECT set on all packets */
805#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700806
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700807struct tcp_congestion_ops {
808 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100809 u32 key;
810 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700811
812 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300813 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700814 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300815 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700816
817 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300818 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700819 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700820 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700821 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300822 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700823 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300824 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +0200825 /* call when ack arrives (optional) */
826 void (*in_ack_event)(struct sock *sk, u32 flags);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700827 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300828 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700829 /* hook for packet ack accounting (optional) */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700830 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300831 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300832 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700833
834 char name[TCP_CA_NAME_MAX];
835 struct module *owner;
836};
837
Joe Perches5c9f3022013-09-23 11:33:32 -0700838int tcp_register_congestion_control(struct tcp_congestion_ops *type);
839void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700840
Florian Westphal55d86942014-09-26 22:37:32 +0200841void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700842void tcp_init_congestion_control(struct sock *sk);
843void tcp_cleanup_congestion_control(struct sock *sk);
844int tcp_set_default_congestion_control(const char *name);
845void tcp_get_default_congestion_control(char *name);
846void tcp_get_available_congestion_control(char *buf, size_t len);
847void tcp_get_allowed_congestion_control(char *buf, size_t len);
848int tcp_set_allowed_congestion_control(char *allowed);
849int tcp_set_congestion_control(struct sock *sk, const char *name);
Neal Cardwelle73ebb02015-01-28 20:01:35 -0500850u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
851void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700852
Joe Perches5c9f3022013-09-23 11:33:32 -0700853u32 tcp_reno_ssthresh(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -0700854void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -0700855extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700856
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100857struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
858u32 tcp_ca_get_key_by_name(const char *name);
Daniel Borkmannea697632015-01-05 23:57:47 +0100859#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100860char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +0100861#else
862static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
863{
864 return NULL;
865}
866#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100867
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200868static inline bool tcp_ca_needs_ecn(const struct sock *sk)
869{
870 const struct inet_connection_sock *icsk = inet_csk(sk);
871
872 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
873}
874
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300875static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700876{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300877 struct inet_connection_sock *icsk = inet_csk(sk);
878
879 if (icsk->icsk_ca_ops->set_state)
880 icsk->icsk_ca_ops->set_state(sk, ca_state);
881 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700882}
883
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300884static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700885{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300886 const struct inet_connection_sock *icsk = inet_csk(sk);
887
888 if (icsk->icsk_ca_ops->cwnd_event)
889 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700890}
891
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300892/* These functions determine how the current flow behaves in respect of SACK
893 * handling. SACK is negotiated with the peer, and therefore it can vary
894 * between different flows.
895 *
896 * tcp_is_sack - SACK enabled
897 * tcp_is_reno - No SACK
898 * tcp_is_fack - FACK enabled, implies SACK enabled
899 */
900static inline int tcp_is_sack(const struct tcp_sock *tp)
901{
902 return tp->rx_opt.sack_ok;
903}
904
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000905static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300906{
907 return !tcp_is_sack(tp);
908}
909
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000910static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300911{
Vijay Subramanianab562222011-12-20 13:23:24 +0000912 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300913}
914
915static inline void tcp_enable_fack(struct tcp_sock *tp)
916{
Vijay Subramanianab562222011-12-20 13:23:24 +0000917 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300918}
919
Yuchung Chengeed530b2012-05-02 13:30:03 +0000920/* TCP early-retransmit (ER) is similar to but more conservative than
921 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
922 */
923static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
924{
925 tp->do_early_retrans = sysctl_tcp_early_retrans &&
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000926 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
927 sysctl_tcp_reordering == 3;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000928}
929
930static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
931{
932 tp->do_early_retrans = 0;
933}
934
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300935static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
936{
937 return tp->sacked_out + tp->lost_out;
938}
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940/* This determines how many packets are "in the network" to the best
941 * of our knowledge. In many cases it is conservative, but where
942 * detailed information is available from the receiver (via SACK
943 * blocks etc.) we can make more aggressive calculations.
944 *
945 * Use this for decisions involving congestion control, use just
946 * tp->packets_out to determine if the send queue is empty or not.
947 *
948 * Read this equation as:
949 *
950 * "Packets sent once on transmission queue" MINUS
951 * "Packets left network, but not honestly ACKed yet" PLUS
952 * "Packets fast retransmitted"
953 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800954static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300956 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -0700959#define TCP_INFINITE_SSTHRESH 0x7fffffff
960
961static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
962{
963 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
964}
965
Yuchung Cheng684bad12012-09-02 17:38:04 +0000966static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
967{
968 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
969 (1 << inet_csk(sk)->icsk_ca_state);
970}
971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +0000973 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 * ssthresh.
975 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300976static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300978 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400979
Yuchung Cheng684bad12012-09-02 17:38:04 +0000980 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 return tp->snd_ssthresh;
982 else
983 return max(tp->snd_ssthresh,
984 ((tp->snd_cwnd >> 1) +
985 (tp->snd_cwnd >> 2)));
986}
987
Ilpo Järvinenb9c45952007-07-27 16:36:17 +0300988/* Use define here intentionally to get WARN_ON location shown at the caller */
989#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Christoph Paasch5ee2c942014-07-14 16:58:32 +0200991void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700992__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Neal Cardwell6b5a5c02011-11-21 17:15:14 +0000994/* The maximum number of MSS of available cwnd for which TSO defers
995 * sending if not using sysctl_tcp_tso_win_divisor.
996 */
997static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
998{
999 return 3;
1000}
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002/* Slow start with delack produces 3 packets of burst, so that
John Heffnerdd9e0dd2008-04-15 15:26:39 -07001003 * it is safe "de facto". This will be the default - same as
1004 * the default reordering threshold - but if reordering increases,
1005 * we must be able to allow cwnd to burst at least this much in order
1006 * to not pull it back when holes are filled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 */
1008static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1009{
John Heffnerdd9e0dd2008-04-15 15:26:39 -07001010 return tp->reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011}
1012
Ilpo Järvinen90840de2007-12-31 04:48:41 -08001013/* Returns end sequence number of the receiver's advertised window */
1014static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1015{
1016 return tp->snd_una + tp->snd_wnd;
1017}
Eric Dumazete114a712014-04-30 11:58:13 -07001018
1019/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1020 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001021 * it was fully used previously. And that's exactly what we do in
1022 * congestion avoidance mode. But in slow start we allow cwnd to grow
1023 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001024 * Example :
1025 * cwnd is 10 (IW10), but application sends 9 frames.
1026 * We allow cwnd to reach 18 when all frames are ACKed.
1027 * This check is safe because it's as aggressive as slow start which already
1028 * risks 100% overshoot. The advantage is that we discourage application to
1029 * either send more filler packets or data to artificially blow up the cwnd
1030 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001031 */
Eric Dumazet24901552014-05-02 21:18:05 -07001032static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001033{
1034 const struct tcp_sock *tp = tcp_sk(sk);
1035
Neal Cardwellca8a2262014-05-22 10:41:08 -04001036 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1037 if (tp->snd_cwnd <= tp->snd_ssthresh)
1038 return tp->snd_cwnd < 2 * tp->max_packets_out;
1039
1040 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001041}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001042
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001043static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001045 const struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001046 const struct inet_connection_sock *icsk = inet_csk(sk);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001047
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001048 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001049 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1050 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051}
1052
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001053static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
1055 tp->snd_wl1 = seq;
1056}
1057
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001058static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
1060 tp->snd_wl1 = seq;
1061}
1062
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063/*
1064 * Calculate(/check) TCP checksum
1065 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001066static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1067 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068{
1069 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1070}
1071
Al Virob51655b2006-11-14 21:40:42 -08001072static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Herbert Xufb286bb2005-11-10 13:01:24 -08001074 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001077static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078{
Herbert Xu60476372007-04-09 11:59:39 -07001079 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 __tcp_checksum_complete(skb);
1081}
1082
1083/* Prequeue for VJ style copy to user, combined with checksumming. */
1084
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001085static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086{
1087 tp->ucopy.task = NULL;
1088 tp->ucopy.len = 0;
1089 tp->ucopy.memory = 0;
1090 skb_queue_head_init(&tp->ucopy.prequeue);
1091}
1092
Joe Perches5c9f3022013-09-23 11:33:32 -07001093bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095#undef STATE_TRACE
1096
1097#ifdef STATE_TRACE
1098static const char *statename[]={
1099 "Unused","Established","Syn Sent","Syn Recv",
1100 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1101 "Close Wait","Last ACK","Listen","Closing"
1102};
1103#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001104void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Joe Perches5c9f3022013-09-23 11:33:32 -07001106void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001108static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
1110 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 rx_opt->num_sacks = 0;
1112}
1113
Joe Perches5c9f3022013-09-23 11:33:32 -07001114u32 tcp_default_init_rwnd(u32 mss);
Yuchung Cheng85f16522013-06-11 15:35:32 -07001115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001117void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1118 __u32 *window_clamp, int wscale_ok,
1119 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121static inline int tcp_win_from_space(int space)
1122{
1123 return sysctl_tcp_adv_win_scale<=0 ?
1124 (space>>(-sysctl_tcp_adv_win_scale)) :
1125 space - (space>>sysctl_tcp_adv_win_scale);
1126}
1127
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001128/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129static inline int tcp_space(const struct sock *sk)
1130{
1131 return tcp_win_from_space(sk->sk_rcvbuf -
1132 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001133}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135static inline int tcp_full_space(const struct sock *sk)
1136{
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001137 return tcp_win_from_space(sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001140extern void tcp_openreq_init_rwin(struct request_sock *req,
1141 struct sock *sk, struct dst_entry *dst);
1142
Joe Perches5c9f3022013-09-23 11:33:32 -07001143void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1146{
1147 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1148}
1149
1150static inline int keepalive_time_when(const struct tcp_sock *tp)
1151{
1152 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1153}
1154
Eric Dumazetdf19a622009-08-28 23:48:54 -07001155static inline int keepalive_probes(const struct tcp_sock *tp)
1156{
1157 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1158}
1159
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001160static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1161{
1162 const struct inet_connection_sock *icsk = &tp->inet_conn;
1163
1164 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1165 tcp_time_stamp - tp->rcv_tstamp);
1166}
1167
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001168static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001170 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1171 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001173 if (fin_timeout < (rto << 2) - (rto >> 1))
1174 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176 return fin_timeout;
1177}
1178
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001179static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1180 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001182 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001183 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001184 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001185 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001186 /*
1187 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1188 * then following tcp messages have valid values. Ignore 0 value,
1189 * or else 'negative' tsval might forbid us to accept their packets.
1190 */
1191 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001192 return true;
1193 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001194}
1195
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001196static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1197 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001198{
1199 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001200 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 /* RST segments are not recommended to carry timestamp,
1203 and, if they do, it is recommended to ignore PAWS because
1204 "their cleanup function should take precedence over timestamps."
1205 Certainly, it is mistake. It is necessary to understand the reasons
1206 of this constraint to relax it: if peer reboots, clock may go
1207 out-of-sync and half-open connections will not be reset.
1208 Actually, the problem would be not existing if all
1209 the implementations followed draft about maintaining clock
1210 via reboots. Linux-2.2 DOES NOT!
1211
1212 However, we can relax time bounds for RST segments to MSL.
1213 */
James Morris9d729f72007-03-04 16:12:44 -08001214 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001215 return false;
1216 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001219bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1220 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001221
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001222static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223{
1224 /* See RFC 2012 */
Pavel Emelyanovcf1100a2008-07-16 20:27:38 -07001225 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1226 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1227 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1228 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
1230
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001231/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001232static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001233{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001234 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001235}
1236
1237static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1238{
1239 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001240 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001241}
1242
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001243/* MD5 Signature */
1244struct crypto_hash;
1245
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001246union tcp_md5_addr {
1247 struct in_addr a4;
1248#if IS_ENABLED(CONFIG_IPV6)
1249 struct in6_addr a6;
1250#endif
1251};
1252
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001253/* - key database */
1254struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001255 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001256 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001257 u8 family; /* AF_INET or AF_INET6 */
1258 union tcp_md5_addr addr;
1259 u8 key[TCP_MD5SIG_MAXKEYLEN];
1260 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001261};
1262
1263/* - sock block */
1264struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001265 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001266 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001267};
1268
1269/* - pseudo header */
1270struct tcp4_pseudohdr {
1271 __be32 saddr;
1272 __be32 daddr;
1273 __u8 pad;
1274 __u8 protocol;
1275 __be16 len;
1276};
1277
1278struct tcp6_pseudohdr {
1279 struct in6_addr saddr;
1280 struct in6_addr daddr;
1281 __be32 len;
1282 __be32 protocol; /* including padding */
1283};
1284
1285union tcp_md5sum_block {
1286 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001287#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001288 struct tcp6_pseudohdr ip6;
1289#endif
1290};
1291
1292/* - pool: digest algorithm, hash description and scratch buffer */
1293struct tcp_md5sig_pool {
1294 struct hash_desc md5_desc;
1295 union tcp_md5sum_block md5_blk;
1296};
1297
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001298/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001299int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1300 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001301int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1302 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1303int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1304 int family);
1305struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001306 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001307
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001308#ifdef CONFIG_TCP_MD5SIG
Joe Perches5c9f3022013-09-23 11:33:32 -07001309struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1310 const union tcp_md5_addr *addr,
1311 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001312#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001313#else
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001314static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1315 const union tcp_md5_addr *addr,
1316 int family)
1317{
1318 return NULL;
1319}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001320#define tcp_twsk_md5_key(twsk) NULL
1321#endif
1322
Joe Perches5c9f3022013-09-23 11:33:32 -07001323bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001324
Joe Perches5c9f3022013-09-23 11:33:32 -07001325struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001326static inline void tcp_put_md5sig_pool(void)
1327{
1328 local_bh_enable();
1329}
Eric Dumazet35790c02010-05-16 00:34:04 -07001330
Joe Perches5c9f3022013-09-23 11:33:32 -07001331int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1332int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1333 unsigned int header_len);
1334int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1335 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001336
Jerry Chu10467162012-08-31 12:29:11 +00001337/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001338void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1339 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1340 unsigned long *last_syn_loss);
1341void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1342 struct tcp_fastopen_cookie *cookie, bool syn_lost);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001343struct tcp_fastopen_request {
1344 /* Fast Open cookie. Size 0 means a cookie request */
1345 struct tcp_fastopen_cookie cookie;
1346 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001347 size_t size;
1348 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001349};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001350void tcp_free_fastopen_req(struct tcp_sock *tp);
1351
Jerry Chu10467162012-08-31 12:29:11 +00001352extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1353int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001354bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1355 struct request_sock *req,
1356 struct tcp_fastopen_cookie *foc,
1357 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001358void tcp_fastopen_init_key_once(bool publish);
Jerry Chu10467162012-08-31 12:29:11 +00001359#define TCP_FASTOPEN_KEY_LENGTH 16
1360
1361/* Fastopen key context */
1362struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001363 struct crypto_cipher *tfm;
1364 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1365 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001366};
1367
David S. Millerfe067e82007-03-07 12:12:44 -08001368/* write queue abstraction */
1369static inline void tcp_write_queue_purge(struct sock *sk)
1370{
1371 struct sk_buff *skb;
1372
1373 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001374 sk_wmem_free_skb(sk, skb);
1375 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001376 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001377}
1378
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001379static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001380{
David S. Millercd07a8e2008-09-23 00:50:13 -07001381 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001382}
1383
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001384static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001385{
David S. Millercd07a8e2008-09-23 00:50:13 -07001386 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001387}
1388
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001389static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1390 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001391{
David S. Millercd07a8e2008-09-23 00:50:13 -07001392 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001393}
1394
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001395static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1396 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001397{
1398 return skb_queue_prev(&sk->sk_write_queue, skb);
1399}
1400
David S. Millerfe067e82007-03-07 12:12:44 -08001401#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001402 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001403
1404#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001405 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001406
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001407#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001408 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001409
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001410static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001411{
1412 return sk->sk_send_head;
1413}
1414
David S. Millercd07a8e2008-09-23 00:50:13 -07001415static inline bool tcp_skb_is_last(const struct sock *sk,
1416 const struct sk_buff *skb)
1417{
1418 return skb_queue_is_last(&sk->sk_write_queue, skb);
1419}
1420
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001421static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001422{
David S. Millercd07a8e2008-09-23 00:50:13 -07001423 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001424 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001425 else
1426 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001427}
1428
1429static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1430{
1431 if (sk->sk_send_head == skb_unlinked)
1432 sk->sk_send_head = NULL;
1433}
1434
1435static inline void tcp_init_send_head(struct sock *sk)
1436{
1437 sk->sk_send_head = NULL;
1438}
1439
1440static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1441{
1442 __skb_queue_tail(&sk->sk_write_queue, skb);
1443}
1444
1445static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1446{
1447 __tcp_add_write_queue_tail(sk, skb);
1448
1449 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001450 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001451 sk->sk_send_head = skb;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001452
1453 if (tcp_sk(sk)->highest_sack == NULL)
1454 tcp_sk(sk)->highest_sack = skb;
1455 }
David S. Millerfe067e82007-03-07 12:12:44 -08001456}
1457
1458static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1459{
1460 __skb_queue_head(&sk->sk_write_queue, skb);
1461}
1462
1463/* Insert buff after skb on the write queue of sk. */
1464static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1465 struct sk_buff *buff,
1466 struct sock *sk)
1467{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001468 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001469}
1470
David S. Miller43f59c82008-09-21 21:28:51 -07001471/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001472static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1473 struct sk_buff *skb,
1474 struct sock *sk)
1475{
David S. Miller43f59c82008-09-21 21:28:51 -07001476 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001477
1478 if (sk->sk_send_head == skb)
1479 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001480}
1481
1482static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1483{
1484 __skb_unlink(skb, &sk->sk_write_queue);
1485}
1486
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001487static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001488{
1489 return skb_queue_empty(&sk->sk_write_queue);
1490}
1491
Krishna Kumar12d50c42009-12-08 22:26:13 +00001492static inline void tcp_push_pending_frames(struct sock *sk)
1493{
1494 if (tcp_send_head(sk)) {
1495 struct tcp_sock *tp = tcp_sk(sk);
1496
1497 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1498 }
1499}
1500
Neal Cardwellecb97192012-02-27 17:52:52 -05001501/* Start sequence of the skb just after the highest skb with SACKed
1502 * bit, valid only if sacked_out > 0 or when the caller has ensured
1503 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001504 */
1505static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1506{
1507 if (!tp->sacked_out)
1508 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001509
1510 if (tp->highest_sack == NULL)
1511 return tp->snd_nxt;
1512
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001513 return TCP_SKB_CB(tp->highest_sack)->seq;
1514}
1515
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001516static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1517{
1518 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1519 tcp_write_queue_next(sk, skb);
1520}
1521
1522static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1523{
1524 return tcp_sk(sk)->highest_sack;
1525}
1526
1527static inline void tcp_highest_sack_reset(struct sock *sk)
1528{
1529 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1530}
1531
1532/* Called when old skb is about to be deleted (to be combined with new skb) */
1533static inline void tcp_highest_sack_combine(struct sock *sk,
1534 struct sk_buff *old,
1535 struct sk_buff *new)
1536{
1537 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1538 tcp_sk(sk)->highest_sack = new;
1539}
1540
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001541/* Determines whether this is a thin stream (which may suffer from
1542 * increased latency). Used to trigger latency-reducing mechanisms.
1543 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001544static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001545{
1546 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1547}
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549/* /proc */
1550enum tcp_seq_states {
1551 TCP_SEQ_STATE_LISTENING,
1552 TCP_SEQ_STATE_OPENREQ,
1553 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554};
1555
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001556int tcp_seq_open(struct inode *inode, struct file *file);
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001559 char *name;
1560 sa_family_t family;
1561 const struct file_operations *seq_fops;
1562 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563};
1564
1565struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001566 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 sa_family_t family;
1568 enum tcp_seq_states state;
1569 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001570 int bucket, offset, sbucket, num;
1571 kuid_t uid;
Tom Herberta8b690f2010-06-07 00:43:42 -07001572 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573};
1574
Joe Perches5c9f3022013-09-23 11:33:32 -07001575int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1576void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001578extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001579extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001580
Joe Perches5c9f3022013-09-23 11:33:32 -07001581void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001582
Eric Dumazet28be6e02013-10-18 10:36:17 -07001583struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001584 netdev_features_t features);
1585struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1586int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001587
Joe Perches5c9f3022013-09-23 11:33:32 -07001588void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001589
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001590static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1591{
1592 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1593}
1594
1595static inline bool tcp_stream_memory_free(const struct sock *sk)
1596{
1597 const struct tcp_sock *tp = tcp_sk(sk);
1598 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1599
1600 return notsent_bytes < tcp_notsent_lowat(tp);
1601}
1602
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001603#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001604int tcp4_proc_init(void);
1605void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001606#endif
1607
Octavian Purdila5db92c92014-06-25 17:09:59 +03001608int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001609int tcp_conn_request(struct request_sock_ops *rsk_ops,
1610 const struct tcp_request_sock_ops *af_ops,
1611 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001612
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001613/* TCP af-specific functions */
1614struct tcp_sock_af_ops {
1615#ifdef CONFIG_TCP_MD5SIG
1616 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001617 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001618 int (*calc_md5_hash)(char *location,
1619 const struct tcp_md5sig_key *md5,
1620 const struct sock *sk,
1621 const struct sk_buff *skb);
1622 int (*md5_parse)(struct sock *sk,
1623 char __user *optval,
1624 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001625#endif
1626};
1627
1628struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001629 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001630#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001631 struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk,
1632 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001633 int (*calc_md5_hash) (char *location,
1634 const struct tcp_md5sig_key *md5,
1635 const struct sock *sk,
1636 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001637#endif
Octavian Purdila16bea702014-06-25 17:09:53 +03001638 void (*init_req)(struct request_sock *req, struct sock *sk,
1639 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001640#ifdef CONFIG_SYN_COOKIES
1641 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1642 __u16 *mss);
1643#endif
Octavian Purdilad94e0412014-06-25 17:09:55 +03001644 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1645 const struct request_sock *req,
1646 bool *strict);
Octavian Purdila936b8bd2014-06-25 17:09:57 +03001647 __u32 (*init_seq)(const struct sk_buff *skb);
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001648 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1649 struct flowi *fl, struct request_sock *req,
1650 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
Octavian Purdila695da142014-06-25 17:10:01 +03001651 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1652 const unsigned long timeout);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001653};
1654
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001655#ifdef CONFIG_SYN_COOKIES
1656static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1657 struct sock *sk, struct sk_buff *skb,
1658 __u16 *mss)
1659{
1660 return ops->cookie_init_seq(sk, skb, mss);
1661}
1662#else
1663static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1664 struct sock *sk, struct sk_buff *skb,
1665 __u16 *mss)
1666{
1667 return 0;
1668}
1669#endif
1670
Joe Perches5c9f3022013-09-23 11:33:32 -07001671int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001672
Joe Perches5c9f3022013-09-23 11:33:32 -07001673void tcp_v4_init(void);
1674void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001675
Cong Wange25f8662014-10-15 14:33:21 -07001676/*
1677 * Save and compile IPv4 options, return a pointer to it
1678 */
1679static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1680{
1681 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1682 struct ip_options_rcu *dopt = NULL;
1683
Cong Wang461b74c2014-10-15 14:33:22 -07001684 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07001685 int opt_size = sizeof(*dopt) + opt->optlen;
1686
1687 dopt = kmalloc(opt_size, GFP_ATOMIC);
1688 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1689 kfree(dopt);
1690 dopt = NULL;
1691 }
1692 }
1693 return dopt;
1694}
1695
Eric Dumazet98781962015-02-03 18:31:53 -08001696/* locally generated TCP pure ACKs have skb->truesize == 2
1697 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1698 * This is much faster than dissecting the packet to find out.
1699 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1700 */
1701static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1702{
1703 return skb->truesize == 2;
1704}
1705
1706static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1707{
1708 skb->truesize = 2;
1709}
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711#endif /* _TCP_H */