blob: d9df80deba31a3266b167d1894e7ebca157e6d44 [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_syn_retries;
244extern int sysctl_tcp_synack_retries;
245extern int sysctl_tcp_retries1;
246extern int sysctl_tcp_retries2;
247extern int sysctl_tcp_orphan_retries;
248extern int sysctl_tcp_syncookies;
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000249extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250extern int sysctl_tcp_retrans_collapse;
251extern int sysctl_tcp_stdurg;
252extern int sysctl_tcp_rfc1337;
253extern int sysctl_tcp_abort_on_overflow;
254extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255extern int sysctl_tcp_fack;
256extern int sysctl_tcp_reordering;
Eric Dumazetdca145f2014-10-27 21:45:24 -0700257extern int sysctl_tcp_max_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700259extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260extern int sysctl_tcp_wmem[3];
261extern int sysctl_tcp_rmem[3];
262extern int sysctl_tcp_app_win;
263extern int sysctl_tcp_adv_win_scale;
264extern int sysctl_tcp_tw_reuse;
265extern int sysctl_tcp_frto;
266extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268extern int sysctl_tcp_moderate_rcvbuf;
269extern int sysctl_tcp_tso_win_divisor;
Rick Jones15d99e02006-03-20 22:40:29 -0800270extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700271extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000272extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000273extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000274extern int sysctl_tcp_early_retrans;
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000275extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200276extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -0700277extern unsigned int sysctl_tcp_notsent_lowat;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700278extern int sysctl_tcp_min_tso_segs;
Yuchung Chengf6722582015-10-16 21:57:42 -0700279extern int sysctl_tcp_min_rtt_wlen;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800280extern int sysctl_tcp_autocorking;
Neal Cardwell032ee422015-02-06 16:04:38 -0500281extern int sysctl_tcp_invalid_ratelimit;
Eric Dumazet43e122b2015-08-21 17:38:02 -0700282extern int sysctl_tcp_pacing_ss_ratio;
283extern int sysctl_tcp_pacing_ca_ratio;
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
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700289/* optimized version of sk_under_memory_pressure() for TCP sockets */
290static inline bool tcp_under_memory_pressure(const struct sock *sk)
291{
Johannes Weinere8056052016-01-14 15:21:14 -0800292 if (mem_cgroup_sockets_enabled && sk->sk_cgrp &&
293 mem_cgroup_under_socket_pressure(sk->sk_cgrp))
294 return true;
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700295
296 return tcp_memory_pressure;
297}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 * The next routines deal with comparing 32 bit unsigned ints
300 * and worry about wraparound (automatic with unsigned arithmetic).
301 */
302
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000303static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800305 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800307#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
309/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000310static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
312 return seq3 - seq2 >= seq1 - seq2;
313}
314
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800315static inline bool tcp_out_of_memory(struct sock *sk)
316{
317 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
318 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
319 return true;
320 return false;
321}
322
Eric Dumazeta6c5ea42015-05-15 12:39:26 -0700323void sk_forced_mem_schedule(struct sock *sk, int size);
324
David S. Millerad1af0f2010-08-25 02:27:49 -0700325static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700326{
David S. Millerad1af0f2010-08-25 02:27:49 -0700327 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
328 int orphans = percpu_counter_read_positive(ocp);
329
330 if (orphans << shift > sysctl_tcp_max_orphans) {
331 orphans = percpu_counter_sum_positive(ocp);
332 if (orphans << shift > sysctl_tcp_max_orphans)
333 return true;
334 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700335 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700336}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Joe Perches5c9f3022013-09-23 11:33:32 -0700338bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800339
Florian Westphala0f82f62009-04-19 09:43:48 +0000340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341extern struct proto tcp_prot;
342
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700343#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
344#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
345#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
346#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 +0000347#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Joe Perches5c9f3022013-09-23 11:33:32 -0700349void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000350
Joe Perches5c9f3022013-09-23 11:33:32 -0700351void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Joe Perches5c9f3022013-09-23 11:33:32 -0700353void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
Joe Perches5c9f3022013-09-23 11:33:32 -0700355void tcp_v4_early_demux(struct sk_buff *skb);
356int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Joe Perches5c9f3022013-09-23 11:33:32 -0700358int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800359int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700360int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
361 int flags);
362void tcp_release_cb(struct sock *sk);
363void tcp_wfree(struct sk_buff *skb);
364void tcp_write_timer_handler(struct sock *sk);
365void tcp_delack_timer_handler(struct sock *sk);
366int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700367int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700368void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
369 const struct tcphdr *th, unsigned int len);
370void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700371int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
372void tcp_twsk_destructor(struct sock *sk);
373ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
374 struct pipe_inode_info *pipe, size_t len,
375 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800376
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700377static inline void tcp_dec_quickack_mode(struct sock *sk,
378 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700380 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700381
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700382 if (icsk->icsk_ack.quick) {
383 if (pkts >= icsk->icsk_ack.quick) {
384 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700385 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700386 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700387 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700388 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 }
390}
391
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700392#define TCP_ECN_OK 1
393#define TCP_ECN_QUEUE_CWR 2
394#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000395#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700396
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000397enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 TCP_TW_SUCCESS = 0,
399 TCP_TW_RST = 1,
400 TCP_TW_ACK = 2,
401 TCP_TW_SYN = 3
402};
403
404
Joe Perches5c9f3022013-09-23 11:33:32 -0700405enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
406 struct sk_buff *skb,
407 const struct tcphdr *th);
408struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazet52452c52015-03-19 19:04:19 -0700409 struct request_sock *req, bool fastopen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700410int tcp_child_process(struct sock *parent, struct sock *child,
411 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400412void tcp_enter_loss(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700413void tcp_clear_retrans(struct tcp_sock *tp);
414void tcp_update_metrics(struct sock *sk);
415void tcp_init_metrics(struct sock *sk);
416void tcp_metrics_init(void);
417bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200418 bool paws_check, bool timestamps);
Joe Perches5c9f3022013-09-23 11:33:32 -0700419bool tcp_remember_stamp(struct sock *sk);
420bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
421void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
422void tcp_disable_fack(struct tcp_sock *tp);
423void tcp_close(struct sock *sk, long timeout);
424void tcp_init_sock(struct sock *sk);
425unsigned int tcp_poll(struct file *file, struct socket *sock,
426 struct poll_table_struct *wait);
427int tcp_getsockopt(struct sock *sk, int level, int optname,
428 char __user *optval, int __user *optlen);
429int tcp_setsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, unsigned int optlen);
431int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000432 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700433int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000434 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700435void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700436void tcp_syn_ack_timeout(const struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800437int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
438 int flags, int *addr_len);
Joe Perches5c9f3022013-09-23 11:33:32 -0700439void tcp_parse_options(const struct sk_buff *skb,
440 struct tcp_options_received *opt_rx,
441 int estab, struct tcp_fastopen_cookie *foc);
442const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900443
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444/*
445 * TCP v4 functions exported for the inet6 API
446 */
447
Joe Perches5c9f3022013-09-23 11:33:32 -0700448void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400449void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet26e37362015-03-22 10:22:22 -0700450void tcp_req_err(struct sock *sk, u32 seq);
Joe Perches5c9f3022013-09-23 11:33:32 -0700451int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700452struct sock *tcp_create_openreq_child(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -0700453 struct request_sock *req,
454 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100455void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Eric Dumazet0c271712015-09-29 07:42:48 -0700456struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700457 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700458 struct dst_entry *dst,
459 struct request_sock *req_unhash,
460 bool *own_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700461int 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);
Eric Dumazet5d062de2015-09-25 07:39:19 -0700464struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
Joe Perches5c9f3022013-09-23 11:33:32 -0700465 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700466 struct tcp_fastopen_cookie *foc,
467 bool attach_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700468int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000470void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000471int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000472void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/* From syncookies.c */
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700475struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
476 struct request_sock *req,
477 struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700478int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
479 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700480struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400481#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200482
Eric Dumazet63262312014-03-19 21:02:21 -0700483/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200484 * This counter is used both as a hash input and partially encoded into
485 * the cookie value. A cookie is only validated further if the delta
486 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200488 * the counter advances immediately after a cookie is generated).
489 */
Eric Dumazet264ea102015-05-14 14:26:56 -0700490#define MAX_SYNCOOKIE_AGE 2
491#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
492#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
493
494/* syncookies: remember time of last synqueue overflow
495 * But do not dirty this field too often (once per second is enough)
Eric Dumazet3f684b42015-09-29 07:42:49 -0700496 * It is racy as we do not hold a lock, but race is very minor.
Eric Dumazet264ea102015-05-14 14:26:56 -0700497 */
Eric Dumazet3f684b42015-09-29 07:42:49 -0700498static inline void tcp_synq_overflow(const struct sock *sk)
Eric Dumazet264ea102015-05-14 14:26:56 -0700499{
500 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
501 unsigned long now = jiffies;
502
503 if (time_after(now, last_overflow + HZ))
504 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
505}
506
507/* syncookies: no recent synqueue overflow on this listening socket? */
508static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
509{
510 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
511
512 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
513}
Florian Westphal8c27bd72013-09-20 22:32:55 +0200514
515static inline u32 tcp_cookie_time(void)
516{
Eric Dumazet63262312014-03-19 21:02:21 -0700517 u64 val = get_jiffies_64();
518
Eric Dumazet264ea102015-05-14 14:26:56 -0700519 do_div(val, TCP_SYNCOOKIE_PERIOD);
Eric Dumazet63262312014-03-19 21:02:21 -0700520 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200521}
522
Joe Perches5c9f3022013-09-23 11:33:32 -0700523u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
524 u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700525__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
Joe Perches5c9f3022013-09-23 11:33:32 -0700526__u32 cookie_init_timestamp(struct request_sock *req);
Florian Westphalf1673382014-11-03 17:35:02 +0100527bool cookie_timestamp_decode(struct tcp_options_received *opt);
528bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100529 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700530
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800531/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700532int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
533 u32 cookie);
534struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100535
Joe Perches5c9f3022013-09-23 11:33:32 -0700536u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
537 const struct tcphdr *th, u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700538__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400539#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540/* tcp_output.c */
541
Joe Perches5c9f3022013-09-23 11:33:32 -0700542void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
543 int nonagle);
544bool tcp_may_send_now(struct sock *sk);
545int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
546int tcp_retransmit_skb(struct sock *, struct sk_buff *);
547void tcp_retransmit_timer(struct sock *sk);
548void tcp_xmit_retransmit_queue(struct sock *);
549void tcp_simple_retransmit(struct sock *);
550int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300551int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Joe Perches5c9f3022013-09-23 11:33:32 -0700553void tcp_send_probe0(struct sock *);
554void tcp_send_partial(struct sock *);
Eric Dumazete520af42015-05-06 14:26:25 -0700555int tcp_write_wakeup(struct sock *, int mib);
Joe Perches5c9f3022013-09-23 11:33:32 -0700556void tcp_send_fin(struct sock *sk);
557void tcp_send_active_reset(struct sock *sk, gfp_t priority);
558int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700559void tcp_push_one(struct sock *, unsigned int mss_now);
560void tcp_send_ack(struct sock *sk);
561void tcp_send_delayed_ack(struct sock *sk);
562void tcp_send_loss_probe(struct sock *sk);
563bool tcp_schedule_loss_probe(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
David S. Millera762a982005-07-05 15:18:51 -0700565/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700566void tcp_resume_early_retransmit(struct sock *sk);
567void tcp_rearm_rto(struct sock *sk);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700568void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700569void tcp_reset(struct sock *sk);
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -0700570void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
David S. Millera762a982005-07-05 15:18:51 -0700571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700573void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700574static inline void tcp_clear_xmit_timers(struct sock *sk)
575{
576 inet_csk_clear_xmit_timers(sk);
577}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Joe Perches5c9f3022013-09-23 11:33:32 -0700579unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
580unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000581
582/* Bound MSS / TSO packet size with the half of the window */
583static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
584{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700585 int cutoff;
586
587 /* When peer uses tiny windows, there is no use in packetizing
588 * to sub-MSS pieces for the sake of SWS or making sure there
589 * are enough packets in the pipe for fast recovery.
590 *
591 * On the other hand, for extremely large MSS devices, handling
592 * smaller than MSS windows in this way does make sense.
593 */
594 if (tp->max_window >= 512)
595 cutoff = (tp->max_window >> 1);
596 else
597 cutoff = tp->max_window;
598
599 if (cutoff && pktsize > cutoff)
600 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000601 else
602 return pktsize;
603}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300605/* tcp.c */
Eric Dumazet0df48c22015-04-28 15:28:17 -0700606void tcp_get_info(struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608/* Read 'sendfile()'-style from a TCP socket */
609typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
610 unsigned int, size_t);
Joe Perches5c9f3022013-09-23 11:33:32 -0700611int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
612 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Joe Perches5c9f3022013-09-23 11:33:32 -0700614void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
Joe Perches5c9f3022013-09-23 11:33:32 -0700616int tcp_mtu_to_mss(struct sock *sk, int pmtu);
617int tcp_mss_to_mtu(struct sock *sk, int mss);
618void tcp_mtup_init(struct sock *sk);
619void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800620
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000621static inline void tcp_bound_rto(const struct sock *sk)
622{
623 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
624 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
625}
626
627static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
628{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800629 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000630}
631
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800632static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
634 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
635 ntohl(TCP_FLAG_ACK) |
636 snd_wnd);
637}
638
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800639static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
641 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
642}
643
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700644static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700646 struct tcp_sock *tp = tcp_sk(sk);
647
David S. Millerb03efcf2005-07-08 14:57:23 -0700648 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 tp->rcv_wnd &&
650 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
651 !tp->urg_data)
652 tcp_fast_path_on(tp);
653}
654
Satoru SATOH0c266892009-05-04 11:11:01 -0700655/* Compute the actual rto_min value */
656static inline u32 tcp_rto_min(struct sock *sk)
657{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400658 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700659 u32 rto_min = TCP_RTO_MIN;
660
661 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
662 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
663 return rto_min;
664}
665
Eric Dumazet740b0f12014-02-26 14:02:48 -0800666static inline u32 tcp_rto_min_us(struct sock *sk)
667{
668 return jiffies_to_usecs(tcp_rto_min(sk));
669}
670
Daniel Borkmann81164412015-01-05 23:57:48 +0100671static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
672{
673 return dst_metric_locked(dst, RTAX_CC_ALGO);
674}
675
Yuchung Chengf6722582015-10-16 21:57:42 -0700676/* Minimum RTT in usec. ~0 means not available. */
677static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
678{
679 return tp->rtt_min[0].rtt;
680}
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682/* Compute the actual receive window we are currently advertising.
683 * Rcv_nxt can be after the window if our peer push more data
684 * than the offered window.
685 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800686static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687{
688 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
689
690 if (win < 0)
691 win = 0;
692 return (u32) win;
693}
694
695/* Choose a new window, without checks for shrinking, and without
696 * scaling applied to the result. The caller does these things
697 * if necessary. This is a "raw" window selection.
698 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700699u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Pavel Emelyanovee995282012-04-19 03:40:39 +0000701void tcp_send_window_probe(struct sock *sk);
702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703/* TCP timestamps are only 32-bits, this causes a slight
704 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800705 * of jiffies in the buffer control blocks below. We decided
706 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 * casts with the following macro.
708 */
709#define tcp_time_stamp ((__u32)(jiffies))
710
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700711static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
712{
713 return skb->skb_mstamp.stamp_jiffies;
714}
715
716
Changli Gaoa3433f32010-06-12 14:01:43 +0000717#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
718
719#define TCPHDR_FIN 0x01
720#define TCPHDR_SYN 0x02
721#define TCPHDR_RST 0x04
722#define TCPHDR_PSH 0x08
723#define TCPHDR_ACK 0x10
724#define TCPHDR_URG 0x20
725#define TCPHDR_ECE 0x40
726#define TCPHDR_CWR 0x80
727
Daniel Borkmann49213552015-05-19 21:04:22 +0200728#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
729
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800730/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700731 * TCP per-packet control information to the transmission code.
732 * We also store the host-order sequence numbers in here too.
733 * This is 44 bytes if IPV6 is enabled.
734 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 */
736struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 __u32 seq; /* Starting sequence number */
738 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700739 union {
740 /* Note : tcp_tw_isn is used in input path only
741 * (isn chosen by tcp_timewait_state_process())
742 *
Eric Dumazetf69ad292015-06-11 09:15:18 -0700743 * tcp_gso_segs/size are used in write queue only,
744 * cf tcp_skb_pcount()/tcp_skb_mss()
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700745 */
746 __u32 tcp_tw_isn;
Eric Dumazetf69ad292015-06-11 09:15:18 -0700747 struct {
748 u16 tcp_gso_segs;
749 u16 tcp_gso_size;
750 };
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700751 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400752 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 __u8 sacked; /* State flags for SACK/FACK. */
755#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
756#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
757#define TCPCB_LOST 0x04 /* SKB is lost */
758#define TCPCB_TAGBITS 0x07 /* All tag bits */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400759#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400761#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
762 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000764 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
765 /* 1 byte hole */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700767 union {
768 struct inet_skb_parm h4;
769#if IS_ENABLED(CONFIG_IPV6)
770 struct inet6_skb_parm h6;
771#endif
772 } header; /* For incoming frames */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773};
774
775#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
776
Eric Dumazet870c3152014-10-17 09:17:20 -0700777
Eric Dumazet815afe12014-10-18 08:34:37 -0700778#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700779/* This is the variant of inet6_iif() that must be used by TCP,
780 * as TCP moves IP6CB into a different location in skb->cb[]
781 */
782static inline int tcp_v6_iif(const struct sk_buff *skb)
783{
784 return TCP_SKB_CB(skb)->header.h6.iif;
785}
Eric Dumazet815afe12014-10-18 08:34:37 -0700786#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788/* Due to TSO, an SKB can be composed of multiple actual
789 * packets. To keep these tracked properly, we use this.
790 */
791static inline int tcp_skb_pcount(const struct sk_buff *skb)
792{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700793 return TCP_SKB_CB(skb)->tcp_gso_segs;
794}
795
796static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
797{
798 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
799}
800
801static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
802{
803 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Eric Dumazetf69ad292015-06-11 09:15:18 -0700806/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807static inline int tcp_skb_mss(const struct sk_buff *skb)
808{
Eric Dumazetf69ad292015-06-11 09:15:18 -0700809 return TCP_SKB_CB(skb)->tcp_gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700812/* Events passed to congestion control interface */
813enum tcp_ca_event {
814 CA_EVENT_TX_START, /* first transmit when no packets in flight */
815 CA_EVENT_CWND_RESTART, /* congestion window restart */
816 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700817 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200818 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
819 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
820 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
821 CA_EVENT_NON_DELAYED_ACK,
Florian Westphal7354c8c2014-09-26 22:37:34 +0200822};
823
Florian Westphal98900922014-09-26 22:37:35 +0200824/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200825enum tcp_ca_ack_event_flags {
Florian Westphal98900922014-09-26 22:37:35 +0200826 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
827 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
828 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700829};
830
831/*
832 * Interface for adding new TCP congestion control handlers
833 */
834#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800835#define TCP_CA_MAX 128
836#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
837
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100838#define TCP_CA_UNSPEC 0
839
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200840/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700841#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200842/* Requires ECN/ECT set on all packets */
843#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700844
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700845union tcp_cc_info;
846
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700847struct tcp_congestion_ops {
848 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100849 u32 key;
850 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700851
852 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300853 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700854 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300855 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700856
857 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300858 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700859 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700860 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700861 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300862 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700863 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300864 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +0200865 /* call when ack arrives (optional) */
866 void (*in_ack_event)(struct sock *sk, u32 flags);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700867 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300868 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700869 /* hook for packet ack accounting (optional) */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700870 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300871 /* get info for inet_diag (optional) */
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700872 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
873 union tcp_cc_info *info);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700874
875 char name[TCP_CA_NAME_MAX];
876 struct module *owner;
877};
878
Joe Perches5c9f3022013-09-23 11:33:32 -0700879int tcp_register_congestion_control(struct tcp_congestion_ops *type);
880void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700881
Florian Westphal55d86942014-09-26 22:37:32 +0200882void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700883void tcp_init_congestion_control(struct sock *sk);
884void tcp_cleanup_congestion_control(struct sock *sk);
885int tcp_set_default_congestion_control(const char *name);
886void tcp_get_default_congestion_control(char *name);
887void tcp_get_available_congestion_control(char *buf, size_t len);
888void tcp_get_allowed_congestion_control(char *buf, size_t len);
889int tcp_set_allowed_congestion_control(char *allowed);
890int tcp_set_congestion_control(struct sock *sk, const char *name);
Neal Cardwelle73ebb082015-01-28 20:01:35 -0500891u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
892void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700893
Joe Perches5c9f3022013-09-23 11:33:32 -0700894u32 tcp_reno_ssthresh(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -0700895void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -0700896extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700897
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100898struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
Daniel Borkmannc3a8d942015-08-31 15:58:47 +0200899u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
Daniel Borkmannea697632015-01-05 23:57:47 +0100900#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100901char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +0100902#else
903static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
904{
905 return NULL;
906}
907#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100908
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200909static inline bool tcp_ca_needs_ecn(const struct sock *sk)
910{
911 const struct inet_connection_sock *icsk = inet_csk(sk);
912
913 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
914}
915
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300916static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700917{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300918 struct inet_connection_sock *icsk = inet_csk(sk);
919
920 if (icsk->icsk_ca_ops->set_state)
921 icsk->icsk_ca_ops->set_state(sk, ca_state);
922 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700923}
924
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300925static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700926{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300927 const struct inet_connection_sock *icsk = inet_csk(sk);
928
929 if (icsk->icsk_ca_ops->cwnd_event)
930 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700931}
932
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300933/* These functions determine how the current flow behaves in respect of SACK
934 * handling. SACK is negotiated with the peer, and therefore it can vary
935 * between different flows.
936 *
937 * tcp_is_sack - SACK enabled
938 * tcp_is_reno - No SACK
939 * tcp_is_fack - FACK enabled, implies SACK enabled
940 */
941static inline int tcp_is_sack(const struct tcp_sock *tp)
942{
943 return tp->rx_opt.sack_ok;
944}
945
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000946static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300947{
948 return !tcp_is_sack(tp);
949}
950
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000951static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300952{
Vijay Subramanianab562222011-12-20 13:23:24 +0000953 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300954}
955
956static inline void tcp_enable_fack(struct tcp_sock *tp)
957{
Vijay Subramanianab562222011-12-20 13:23:24 +0000958 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300959}
960
Yuchung Chengeed530b2012-05-02 13:30:03 +0000961/* TCP early-retransmit (ER) is similar to but more conservative than
962 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
963 */
964static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
965{
966 tp->do_early_retrans = sysctl_tcp_early_retrans &&
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000967 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
968 sysctl_tcp_reordering == 3;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000969}
970
971static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
972{
973 tp->do_early_retrans = 0;
974}
975
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300976static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
977{
978 return tp->sacked_out + tp->lost_out;
979}
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/* This determines how many packets are "in the network" to the best
982 * of our knowledge. In many cases it is conservative, but where
983 * detailed information is available from the receiver (via SACK
984 * blocks etc.) we can make more aggressive calculations.
985 *
986 * Use this for decisions involving congestion control, use just
987 * tp->packets_out to determine if the send queue is empty or not.
988 *
989 * Read this equation as:
990 *
991 * "Packets sent once on transmission queue" MINUS
992 * "Packets left network, but not honestly ACKed yet" PLUS
993 * "Packets fast retransmitted"
994 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800995static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300997 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998}
999
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001000#define TCP_INFINITE_SSTHRESH 0x7fffffff
1001
Yuchung Cheng071d5082015-07-09 13:16:29 -07001002static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1003{
Yuchung Cheng76174002015-07-09 13:16:30 -07001004 return tp->snd_cwnd < tp->snd_ssthresh;
Yuchung Cheng071d5082015-07-09 13:16:29 -07001005}
1006
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001007static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1008{
1009 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1010}
1011
Yuchung Cheng684bad12012-09-02 17:38:04 +00001012static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1013{
1014 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1015 (1 << inet_csk(sk)->icsk_ca_state);
1016}
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +00001019 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 * ssthresh.
1021 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001022static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001024 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001025
Yuchung Cheng684bad12012-09-02 17:38:04 +00001026 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 return tp->snd_ssthresh;
1028 else
1029 return max(tp->snd_ssthresh,
1030 ((tp->snd_cwnd >> 1) +
1031 (tp->snd_cwnd >> 2)));
1032}
1033
Ilpo Järvinenb9c45952007-07-27 16:36:17 +03001034/* Use define here intentionally to get WARN_ON location shown at the caller */
1035#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Christoph Paasch5ee2c942014-07-14 16:58:32 +02001037void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001038__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Neal Cardwell6b5a5c02011-11-21 17:15:14 +00001040/* The maximum number of MSS of available cwnd for which TSO defers
1041 * sending if not using sysctl_tcp_tso_win_divisor.
1042 */
1043static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1044{
1045 return 3;
1046}
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048/* Slow start with delack produces 3 packets of burst, so that
John Heffnerdd9e0dd2008-04-15 15:26:39 -07001049 * it is safe "de facto". This will be the default - same as
1050 * the default reordering threshold - but if reordering increases,
1051 * we must be able to allow cwnd to burst at least this much in order
1052 * to not pull it back when holes are filled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 */
1054static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1055{
John Heffnerdd9e0dd2008-04-15 15:26:39 -07001056 return tp->reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057}
1058
Ilpo Järvinen90840de2007-12-31 04:48:41 -08001059/* Returns end sequence number of the receiver's advertised window */
1060static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1061{
1062 return tp->snd_una + tp->snd_wnd;
1063}
Eric Dumazete114a712014-04-30 11:58:13 -07001064
1065/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1066 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001067 * it was fully used previously. And that's exactly what we do in
1068 * congestion avoidance mode. But in slow start we allow cwnd to grow
1069 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001070 * Example :
1071 * cwnd is 10 (IW10), but application sends 9 frames.
1072 * We allow cwnd to reach 18 when all frames are ACKed.
1073 * This check is safe because it's as aggressive as slow start which already
1074 * risks 100% overshoot. The advantage is that we discourage application to
1075 * either send more filler packets or data to artificially blow up the cwnd
1076 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001077 */
Eric Dumazet24901552014-05-02 21:18:05 -07001078static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001079{
1080 const struct tcp_sock *tp = tcp_sk(sk);
1081
Neal Cardwellca8a2262014-05-22 10:41:08 -04001082 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
Yuchung Cheng071d5082015-07-09 13:16:29 -07001083 if (tcp_in_slow_start(tp))
Neal Cardwellca8a2262014-05-22 10:41:08 -04001084 return tp->snd_cwnd < 2 * tp->max_packets_out;
1085
1086 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001087}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001088
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001089/* Something is really bad, we could not queue an additional packet,
1090 * because qdisc is full or receiver sent a 0 window.
1091 * We do not want to add fuel to the fire, or abort too early,
1092 * so make sure the timer we arm now is at least 200ms in the future,
1093 * regardless of current icsk_rto value (as it could be ~2ms)
1094 */
1095static inline unsigned long tcp_probe0_base(const struct sock *sk)
1096{
1097 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1098}
1099
1100/* Variant of inet_csk_rto_backoff() used for zero window probes */
1101static inline unsigned long tcp_probe0_when(const struct sock *sk,
1102 unsigned long max_when)
1103{
1104 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1105
1106 return (unsigned long)min_t(u64, when, max_when);
1107}
1108
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001109static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001111 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001112 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001113 tcp_probe0_base(sk), TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114}
1115
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001116static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
1118 tp->snd_wl1 = seq;
1119}
1120
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001121static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122{
1123 tp->snd_wl1 = seq;
1124}
1125
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126/*
1127 * Calculate(/check) TCP checksum
1128 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001129static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1130 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131{
1132 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1133}
1134
Al Virob51655b2006-11-14 21:40:42 -08001135static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Herbert Xufb286bb2005-11-10 13:01:24 -08001137 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001140static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
Herbert Xu60476372007-04-09 11:59:39 -07001142 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 __tcp_checksum_complete(skb);
1144}
1145
1146/* Prequeue for VJ style copy to user, combined with checksumming. */
1147
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001148static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
1150 tp->ucopy.task = NULL;
1151 tp->ucopy.len = 0;
1152 tp->ucopy.memory = 0;
1153 skb_queue_head_init(&tp->ucopy.prequeue);
1154}
1155
Joe Perches5c9f3022013-09-23 11:33:32 -07001156bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
1158#undef STATE_TRACE
1159
1160#ifdef STATE_TRACE
1161static const char *statename[]={
1162 "Unused","Established","Syn Sent","Syn Recv",
1163 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1164 "Close Wait","Last ACK","Listen","Closing"
1165};
1166#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001167void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Joe Perches5c9f3022013-09-23 11:33:32 -07001169void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09001171int tcp_abort(struct sock *sk, int err);
1172
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001173static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174{
1175 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 rx_opt->num_sacks = 0;
1177}
1178
Joe Perches5c9f3022013-09-23 11:33:32 -07001179u32 tcp_default_init_rwnd(u32 mss);
Eric Dumazet6f021c62015-08-21 12:30:00 -07001180void tcp_cwnd_restart(struct sock *sk, s32 delta);
1181
1182static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1183{
1184 struct tcp_sock *tp = tcp_sk(sk);
1185 s32 delta;
1186
1187 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1188 return;
1189 delta = tcp_time_stamp - tp->lsndtime;
1190 if (delta > inet_csk(sk)->icsk_rto)
1191 tcp_cwnd_restart(sk, delta);
1192}
Yuchung Cheng85f16522013-06-11 15:35:32 -07001193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001195void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1196 __u32 *window_clamp, int wscale_ok,
1197 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199static inline int tcp_win_from_space(int space)
1200{
1201 return sysctl_tcp_adv_win_scale<=0 ?
1202 (space>>(-sysctl_tcp_adv_win_scale)) :
1203 space - (space>>sysctl_tcp_adv_win_scale);
1204}
1205
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001206/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207static inline int tcp_space(const struct sock *sk)
1208{
1209 return tcp_win_from_space(sk->sk_rcvbuf -
1210 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001211}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213static inline int tcp_full_space(const struct sock *sk)
1214{
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001215 return tcp_win_from_space(sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
1217
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001218extern void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -07001219 const struct sock *sk_listener,
1220 const struct dst_entry *dst);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001221
Joe Perches5c9f3022013-09-23 11:33:32 -07001222void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1225{
Nikolay Borisovb840d152016-01-07 16:38:45 +02001226 struct net *net = sock_net((struct sock *)tp);
1227
1228 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
1230
1231static inline int keepalive_time_when(const struct tcp_sock *tp)
1232{
Nikolay Borisov13b287e2016-01-07 16:38:43 +02001233 struct net *net = sock_net((struct sock *)tp);
1234
1235 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236}
1237
Eric Dumazetdf19a622009-08-28 23:48:54 -07001238static inline int keepalive_probes(const struct tcp_sock *tp)
1239{
Nikolay Borisov9bd68612016-01-07 16:38:44 +02001240 struct net *net = sock_net((struct sock *)tp);
1241
1242 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
Eric Dumazetdf19a622009-08-28 23:48:54 -07001243}
1244
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001245static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1246{
1247 const struct inet_connection_sock *icsk = &tp->inet_conn;
1248
1249 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1250 tcp_time_stamp - tp->rcv_tstamp);
1251}
1252
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001253static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001255 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1256 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001258 if (fin_timeout < (rto << 2) - (rto >> 1))
1259 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
1261 return fin_timeout;
1262}
1263
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001264static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1265 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001267 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001268 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001269 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001270 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001271 /*
1272 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1273 * then following tcp messages have valid values. Ignore 0 value,
1274 * or else 'negative' tsval might forbid us to accept their packets.
1275 */
1276 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001277 return true;
1278 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001279}
1280
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001281static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1282 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001283{
1284 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001285 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
1287 /* RST segments are not recommended to carry timestamp,
1288 and, if they do, it is recommended to ignore PAWS because
1289 "their cleanup function should take precedence over timestamps."
1290 Certainly, it is mistake. It is necessary to understand the reasons
1291 of this constraint to relax it: if peer reboots, clock may go
1292 out-of-sync and half-open connections will not be reset.
1293 Actually, the problem would be not existing if all
1294 the implementations followed draft about maintaining clock
1295 via reboots. Linux-2.2 DOES NOT!
1296
1297 However, we can relax time bounds for RST segments to MSL.
1298 */
James Morris9d729f72007-03-04 16:12:44 -08001299 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001300 return false;
1301 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001304bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1305 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001306
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001307static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
1309 /* See RFC 2012 */
Pavel Emelyanovcf1100a2008-07-16 20:27:38 -07001310 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1311 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1312 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1313 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001316/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001317static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001318{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001319 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001320}
1321
1322static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1323{
1324 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001325 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001326}
1327
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001328/* MD5 Signature */
1329struct crypto_hash;
1330
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001331union tcp_md5_addr {
1332 struct in_addr a4;
1333#if IS_ENABLED(CONFIG_IPV6)
1334 struct in6_addr a6;
1335#endif
1336};
1337
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001338/* - key database */
1339struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001340 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001341 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001342 u8 family; /* AF_INET or AF_INET6 */
1343 union tcp_md5_addr addr;
1344 u8 key[TCP_MD5SIG_MAXKEYLEN];
1345 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001346};
1347
1348/* - sock block */
1349struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001350 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001351 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001352};
1353
1354/* - pseudo header */
1355struct tcp4_pseudohdr {
1356 __be32 saddr;
1357 __be32 daddr;
1358 __u8 pad;
1359 __u8 protocol;
1360 __be16 len;
1361};
1362
1363struct tcp6_pseudohdr {
1364 struct in6_addr saddr;
1365 struct in6_addr daddr;
1366 __be32 len;
1367 __be32 protocol; /* including padding */
1368};
1369
1370union tcp_md5sum_block {
1371 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001372#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001373 struct tcp6_pseudohdr ip6;
1374#endif
1375};
1376
1377/* - pool: digest algorithm, hash description and scratch buffer */
1378struct tcp_md5sig_pool {
1379 struct hash_desc md5_desc;
1380 union tcp_md5sum_block md5_blk;
1381};
1382
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001383/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001384int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1385 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001386int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1387 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1388int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1389 int family);
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001390struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001391 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001392
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001393#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001394struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -07001395 const union tcp_md5_addr *addr,
1396 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001397#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001398#else
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001399static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001400 const union tcp_md5_addr *addr,
1401 int family)
1402{
1403 return NULL;
1404}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001405#define tcp_twsk_md5_key(twsk) NULL
1406#endif
1407
Joe Perches5c9f3022013-09-23 11:33:32 -07001408bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001409
Joe Perches5c9f3022013-09-23 11:33:32 -07001410struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001411static inline void tcp_put_md5sig_pool(void)
1412{
1413 local_bh_enable();
1414}
Eric Dumazet35790c02010-05-16 00:34:04 -07001415
Joe Perches5c9f3022013-09-23 11:33:32 -07001416int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1417int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1418 unsigned int header_len);
1419int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1420 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001421
Jerry Chu10467162012-08-31 12:29:11 +00001422/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001423void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1424 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1425 unsigned long *last_syn_loss);
1426void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -07001427 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1428 u16 try_exp);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001429struct tcp_fastopen_request {
1430 /* Fast Open cookie. Size 0 means a cookie request */
1431 struct tcp_fastopen_cookie cookie;
1432 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001433 size_t size;
1434 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001435};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001436void tcp_free_fastopen_req(struct tcp_sock *tp);
1437
Jerry Chu10467162012-08-31 12:29:11 +00001438extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1439int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Eric Dumazet7c85af82015-09-24 17:16:05 -07001440struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1441 struct request_sock *req,
1442 struct tcp_fastopen_cookie *foc,
1443 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001444void tcp_fastopen_init_key_once(bool publish);
Jerry Chu10467162012-08-31 12:29:11 +00001445#define TCP_FASTOPEN_KEY_LENGTH 16
1446
1447/* Fastopen key context */
1448struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001449 struct crypto_cipher *tfm;
1450 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1451 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001452};
1453
David S. Millerfe067e82007-03-07 12:12:44 -08001454/* write queue abstraction */
1455static inline void tcp_write_queue_purge(struct sock *sk)
1456{
1457 struct sk_buff *skb;
1458
1459 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001460 sk_wmem_free_skb(sk, skb);
1461 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001462 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001463}
1464
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001465static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001466{
David S. Millercd07a8e2008-09-23 00:50:13 -07001467 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001468}
1469
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001470static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001471{
David S. Millercd07a8e2008-09-23 00:50:13 -07001472 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001473}
1474
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001475static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1476 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001477{
David S. Millercd07a8e2008-09-23 00:50:13 -07001478 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001479}
1480
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001481static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1482 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001483{
1484 return skb_queue_prev(&sk->sk_write_queue, skb);
1485}
1486
David S. Millerfe067e82007-03-07 12:12:44 -08001487#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001488 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001489
1490#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001491 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001492
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001493#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001494 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001495
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001496static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001497{
1498 return sk->sk_send_head;
1499}
1500
David S. Millercd07a8e2008-09-23 00:50:13 -07001501static inline bool tcp_skb_is_last(const struct sock *sk,
1502 const struct sk_buff *skb)
1503{
1504 return skb_queue_is_last(&sk->sk_write_queue, skb);
1505}
1506
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001507static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001508{
David S. Millercd07a8e2008-09-23 00:50:13 -07001509 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001510 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001511 else
1512 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001513}
1514
1515static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1516{
1517 if (sk->sk_send_head == skb_unlinked)
1518 sk->sk_send_head = NULL;
1519}
1520
1521static inline void tcp_init_send_head(struct sock *sk)
1522{
1523 sk->sk_send_head = NULL;
1524}
1525
1526static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1527{
1528 __skb_queue_tail(&sk->sk_write_queue, skb);
1529}
1530
1531static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1532{
1533 __tcp_add_write_queue_tail(sk, skb);
1534
1535 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001536 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001537 sk->sk_send_head = skb;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001538
1539 if (tcp_sk(sk)->highest_sack == NULL)
1540 tcp_sk(sk)->highest_sack = skb;
1541 }
David S. Millerfe067e82007-03-07 12:12:44 -08001542}
1543
1544static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1545{
1546 __skb_queue_head(&sk->sk_write_queue, skb);
1547}
1548
1549/* Insert buff after skb on the write queue of sk. */
1550static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1551 struct sk_buff *buff,
1552 struct sock *sk)
1553{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001554 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001555}
1556
David S. Miller43f59c82008-09-21 21:28:51 -07001557/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001558static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1559 struct sk_buff *skb,
1560 struct sock *sk)
1561{
David S. Miller43f59c82008-09-21 21:28:51 -07001562 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001563
1564 if (sk->sk_send_head == skb)
1565 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001566}
1567
1568static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1569{
1570 __skb_unlink(skb, &sk->sk_write_queue);
1571}
1572
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001573static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001574{
1575 return skb_queue_empty(&sk->sk_write_queue);
1576}
1577
Krishna Kumar12d50c42009-12-08 22:26:13 +00001578static inline void tcp_push_pending_frames(struct sock *sk)
1579{
1580 if (tcp_send_head(sk)) {
1581 struct tcp_sock *tp = tcp_sk(sk);
1582
1583 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1584 }
1585}
1586
Neal Cardwellecb97192012-02-27 17:52:52 -05001587/* Start sequence of the skb just after the highest skb with SACKed
1588 * bit, valid only if sacked_out > 0 or when the caller has ensured
1589 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001590 */
1591static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1592{
1593 if (!tp->sacked_out)
1594 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001595
1596 if (tp->highest_sack == NULL)
1597 return tp->snd_nxt;
1598
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001599 return TCP_SKB_CB(tp->highest_sack)->seq;
1600}
1601
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001602static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1603{
1604 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1605 tcp_write_queue_next(sk, skb);
1606}
1607
1608static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1609{
1610 return tcp_sk(sk)->highest_sack;
1611}
1612
1613static inline void tcp_highest_sack_reset(struct sock *sk)
1614{
1615 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1616}
1617
1618/* Called when old skb is about to be deleted (to be combined with new skb) */
1619static inline void tcp_highest_sack_combine(struct sock *sk,
1620 struct sk_buff *old,
1621 struct sk_buff *new)
1622{
1623 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1624 tcp_sk(sk)->highest_sack = new;
1625}
1626
Florian Westphalb1f0a0e2015-12-21 21:29:24 +01001627/* This helper checks if socket has IP_TRANSPARENT set */
1628static inline bool inet_sk_transparent(const struct sock *sk)
1629{
1630 switch (sk->sk_state) {
1631 case TCP_TIME_WAIT:
1632 return inet_twsk(sk)->tw_transparent;
1633 case TCP_NEW_SYN_RECV:
1634 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1635 }
1636 return inet_sk(sk)->transparent;
1637}
1638
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001639/* Determines whether this is a thin stream (which may suffer from
1640 * increased latency). Used to trigger latency-reducing mechanisms.
1641 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001642static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001643{
1644 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1645}
1646
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647/* /proc */
1648enum tcp_seq_states {
1649 TCP_SEQ_STATE_LISTENING,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651};
1652
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001653int tcp_seq_open(struct inode *inode, struct file *file);
1654
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001656 char *name;
1657 sa_family_t family;
1658 const struct file_operations *seq_fops;
1659 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660};
1661
1662struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001663 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 sa_family_t family;
1665 enum tcp_seq_states state;
1666 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001667 int bucket, offset, sbucket, num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001668 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669};
1670
Joe Perches5c9f3022013-09-23 11:33:32 -07001671int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1672void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001674extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001675extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001676
Joe Perches5c9f3022013-09-23 11:33:32 -07001677void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001678
Eric Dumazet28be6e02013-10-18 10:36:17 -07001679struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001680 netdev_features_t features);
1681struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1682int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001683
Joe Perches5c9f3022013-09-23 11:33:32 -07001684void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001685
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001686static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1687{
1688 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1689}
1690
1691static inline bool tcp_stream_memory_free(const struct sock *sk)
1692{
1693 const struct tcp_sock *tp = tcp_sk(sk);
1694 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1695
1696 return notsent_bytes < tcp_notsent_lowat(tp);
1697}
1698
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001699#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001700int tcp4_proc_init(void);
1701void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001702#endif
1703
Eric Dumazetea3bea32015-09-25 07:39:23 -07001704int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001705int tcp_conn_request(struct request_sock_ops *rsk_ops,
1706 const struct tcp_request_sock_ops *af_ops,
1707 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001708
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001709/* TCP af-specific functions */
1710struct tcp_sock_af_ops {
1711#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001712 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001713 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001714 int (*calc_md5_hash)(char *location,
1715 const struct tcp_md5sig_key *md5,
1716 const struct sock *sk,
1717 const struct sk_buff *skb);
1718 int (*md5_parse)(struct sock *sk,
1719 char __user *optval,
1720 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001721#endif
1722};
1723
1724struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001725 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001726#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001727 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001728 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001729 int (*calc_md5_hash) (char *location,
1730 const struct tcp_md5sig_key *md5,
1731 const struct sock *sk,
1732 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001733#endif
Eric Dumazetb40cf182015-09-25 07:39:08 -07001734 void (*init_req)(struct request_sock *req,
1735 const struct sock *sk_listener,
Octavian Purdila16bea702014-06-25 17:09:53 +03001736 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001737#ifdef CONFIG_SYN_COOKIES
Eric Dumazet3f684b42015-09-29 07:42:49 -07001738 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001739 __u16 *mss);
1740#endif
Eric Dumazetf9646292015-09-29 07:42:50 -07001741 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
Octavian Purdilad94e0412014-06-25 17:09:55 +03001742 const struct request_sock *req,
1743 bool *strict);
Octavian Purdila936b8bd2014-06-25 17:09:57 +03001744 __u32 (*init_seq)(const struct sk_buff *skb);
Eric Dumazet0f935db2015-09-25 07:39:21 -07001745 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001746 struct flowi *fl, struct request_sock *req,
Eric Dumazetdc6ef6b2015-10-16 13:00:01 -07001747 struct tcp_fastopen_cookie *foc,
Eric Dumazetca6fb062015-10-02 11:43:35 -07001748 bool attach_req);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001749};
1750
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001751#ifdef CONFIG_SYN_COOKIES
1752static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001753 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001754 __u16 *mss)
1755{
Eric Dumazet3f684b42015-09-29 07:42:49 -07001756 tcp_synq_overflow(sk);
1757 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1758 return ops->cookie_init_seq(skb, mss);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001759}
1760#else
1761static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001762 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001763 __u16 *mss)
1764{
1765 return 0;
1766}
1767#endif
1768
Joe Perches5c9f3022013-09-23 11:33:32 -07001769int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001770
Joe Perches5c9f3022013-09-23 11:33:32 -07001771void tcp_v4_init(void);
1772void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001773
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001774/* tcp_recovery.c */
1775
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -07001776/* Flags to enable various loss recovery features. See below */
1777extern int sysctl_tcp_recovery;
1778
1779/* Use TCP RACK to detect (some) tail and retransmit losses */
1780#define TCP_RACK_LOST_RETRANS 0x1
1781
1782extern int tcp_rack_mark_lost(struct sock *sk);
1783
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001784extern void tcp_rack_advance(struct tcp_sock *tp,
1785 const struct skb_mstamp *xmit_time, u8 sacked);
1786
Cong Wange25f8662014-10-15 14:33:21 -07001787/*
1788 * Save and compile IPv4 options, return a pointer to it
1789 */
1790static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1791{
1792 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1793 struct ip_options_rcu *dopt = NULL;
1794
Cong Wang461b74c2014-10-15 14:33:22 -07001795 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07001796 int opt_size = sizeof(*dopt) + opt->optlen;
1797
1798 dopt = kmalloc(opt_size, GFP_ATOMIC);
1799 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1800 kfree(dopt);
1801 dopt = NULL;
1802 }
1803 }
1804 return dopt;
1805}
1806
Eric Dumazet98781962015-02-03 18:31:53 -08001807/* locally generated TCP pure ACKs have skb->truesize == 2
1808 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1809 * This is much faster than dissecting the packet to find out.
1810 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1811 */
1812static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1813{
1814 return skb->truesize == 2;
1815}
1816
1817static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1818{
1819 skb->truesize = 2;
1820}
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822#endif /* _TCP_H */