blob: c81f3b958d44f1cece510426b590be7af33d9b65 [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>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080030#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000031#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080032#include <linux/ktime.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070033
34#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070035#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070036#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070038#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070042#include <net/tcp_states.h>
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -070043#include <net/inet_ecn.h>
Satoru SATOH0c266892009-05-04 11:11:01 -070044#include <net/dst.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/seq_file.h>
Glauber Costa180d8cd2011-12-11 21:47:02 +000047#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070049extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Eric Dumazetdd24c002008-11-25 21:17:14 -080051extern struct percpu_counter tcp_orphan_count;
Joe Perches5c9f3022013-09-23 11:33:32 -070052void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070055#define MAX_TCP_OPTION_SPACE 40
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090057/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090059 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 */
61#define MAX_TCP_WINDOW 32767U
62
63/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64#define TCP_MIN_MSS 88U
65
John Heffner5d424d52006-03-20 17:53:41 -080066/* The least MTU to use for probing */
Fan Dudcd8fb82015-03-06 11:18:22 +080067#define TCP_BASE_MSS 1024
John Heffner5d424d52006-03-20 17:53:41 -080068
Fan Du05cbc0d2015-03-06 11:18:24 +080069/* probing interval, default to 10 minutes as per RFC4821 */
70#define TCP_PROBE_INTERVAL 600
71
Fan Du6b58e0a2015-03-06 11:18:23 +080072/* Specify interval when tcp mtu probing will stop */
73#define TCP_PROBE_THRESHOLD 8
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075/* After receiving this amount of duplicate ACKs fast retransmit starts. */
76#define TCP_FASTRETRANS_THRESH 3
77
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/* Maximal number of ACKs sent quickly to accelerate slow-start. */
79#define TCP_MAX_QUICKACKS 16U
80
81/* urg_data states */
82#define TCP_URG_VALID 0x0100
83#define TCP_URG_NOTYET 0x0200
84#define TCP_URG_READ 0x0400
85
86#define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
91 */
92
93#define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
98 */
99
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000100#define TCP_SYN_RETRIES 6 /* This is how many retries are done
101 * when active opening a connection.
102 * RFC1122 says the minimum retry MUST
103 * be at least 180secs. Nevertheless
104 * this value is corresponding to
105 * 63secs of retransmission with the
106 * current initial RTO.
107 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000109#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
110 * when passive opening a connection.
111 * This is corresponding to 31secs of
112 * retransmission with the current
113 * initial RTO.
114 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
117 * state, about 60 seconds */
118#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
119 /* BSD style FIN_WAIT2 deadlock breaker.
120 * It used to be 3min, new value is 60sec,
121 * to combine FIN-WAIT-2 timeout with
122 * TIME-WAIT timer.
123 */
124
125#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
126#if HZ >= 100
127#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
128#define TCP_ATO_MIN ((unsigned)(HZ/25))
129#else
130#define TCP_DELACK_MIN 4U
131#define TCP_ATO_MIN 4U
132#endif
133#define TCP_RTO_MAX ((unsigned)(120*HZ))
134#define TCP_RTO_MIN ((unsigned)(HZ/5))
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000135#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000136#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
137 * used as a fallback RTO for the
138 * initial data transmission if no
139 * valid RTT sample has been acquired,
140 * most likely due to retrans in 3WHS.
141 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
144 * for local resources.
145 */
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800146#define TCP_REO_TIMEOUT_MIN (2000) /* Min RACK reordering timeout in usec */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
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
Jörg Thalheim21603fc2016-01-27 22:54:06 +0100219/* TCP initial congestion window as per rfc6928 */
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
Yuchung Chengcebc5cb2016-08-22 17:17:54 -0700231 * TCP_FASTOPEN socket option.
Jerry Chu10467162012-08-31 12:29:11 +0000232 */
233#define TFO_SERVER_WO_SOCKOPT1 0x400
Jerry Chu10467162012-08-31 12:29:11 +0000234
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237extern int sysctl_tcp_timestamps;
238extern int sysctl_tcp_window_scaling;
239extern int sysctl_tcp_sack;
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000240extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241extern int sysctl_tcp_retrans_collapse;
242extern int sysctl_tcp_stdurg;
243extern int sysctl_tcp_rfc1337;
244extern int sysctl_tcp_abort_on_overflow;
245extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246extern int sysctl_tcp_fack;
247extern int sysctl_tcp_reordering;
Eric Dumazetdca145f2014-10-27 21:45:24 -0700248extern int sysctl_tcp_max_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700250extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251extern int sysctl_tcp_wmem[3];
252extern int sysctl_tcp_rmem[3];
253extern int sysctl_tcp_app_win;
254extern int sysctl_tcp_adv_win_scale;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255extern int sysctl_tcp_frto;
256extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258extern int sysctl_tcp_moderate_rcvbuf;
259extern int sysctl_tcp_tso_win_divisor;
Rick Jones15d99e02006-03-20 22:40:29 -0800260extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700261extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000262extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000263extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000264extern int sysctl_tcp_early_retrans;
Yuchung Chenga0370b32017-01-12 22:11:36 -0800265extern int sysctl_tcp_recovery;
266#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
267
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000268extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200269extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700270extern int sysctl_tcp_min_tso_segs;
Yuchung Chengf6722582015-10-16 21:57:42 -0700271extern int sysctl_tcp_min_rtt_wlen;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800272extern int sysctl_tcp_autocorking;
Neal Cardwell032ee422015-02-06 16:04:38 -0500273extern int sysctl_tcp_invalid_ratelimit;
Eric Dumazet43e122b2015-08-21 17:38:02 -0700274extern int sysctl_tcp_pacing_ss_ratio;
275extern int sysctl_tcp_pacing_ca_ratio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Eric Dumazet8d987e52010-11-09 23:24:26 +0000277extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800278extern struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279extern int tcp_memory_pressure;
280
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700281/* optimized version of sk_under_memory_pressure() for TCP sockets */
282static inline bool tcp_under_memory_pressure(const struct sock *sk)
283{
Johannes Weinerbaac50b2016-01-14 15:21:17 -0800284 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
285 mem_cgroup_under_socket_pressure(sk->sk_memcg))
Johannes Weinere8056052016-01-14 15:21:14 -0800286 return true;
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700287
288 return tcp_memory_pressure;
289}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
293 */
294
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000295static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800297 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800299#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
301/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000302static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
304 return seq3 - seq2 >= seq1 - seq2;
305}
306
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800307static inline bool tcp_out_of_memory(struct sock *sk)
308{
309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
311 return true;
312 return false;
313}
314
Eric Dumazeta6c5ea42015-05-15 12:39:26 -0700315void sk_forced_mem_schedule(struct sock *sk, int size);
316
David S. Millerad1af0f2010-08-25 02:27:49 -0700317static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700318{
David S. Millerad1af0f2010-08-25 02:27:49 -0700319 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
320 int orphans = percpu_counter_read_positive(ocp);
321
322 if (orphans << shift > sysctl_tcp_max_orphans) {
323 orphans = percpu_counter_sum_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans)
325 return true;
326 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700327 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700328}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Joe Perches5c9f3022013-09-23 11:33:32 -0700330bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800331
Florian Westphala0f82f62009-04-19 09:43:48 +0000332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333extern struct proto tcp_prot;
334
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700335#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700336#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700337#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000338#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Joe Perches5c9f3022013-09-23 11:33:32 -0700340void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000341
Joe Perches5c9f3022013-09-23 11:33:32 -0700342void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Joe Perches5c9f3022013-09-23 11:33:32 -0700344void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Joe Perches5c9f3022013-09-23 11:33:32 -0700346void tcp_v4_early_demux(struct sk_buff *skb);
347int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Joe Perches5c9f3022013-09-23 11:33:32 -0700349int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800350int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700351int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
352 int flags);
353void tcp_release_cb(struct sock *sk);
354void tcp_wfree(struct sk_buff *skb);
355void tcp_write_timer_handler(struct sock *sk);
356void tcp_delack_timer_handler(struct sock *sk);
357int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700358int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700359void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
360 const struct tcphdr *th, unsigned int len);
361void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700362int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
363void tcp_twsk_destructor(struct sock *sk);
364ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
365 struct pipe_inode_info *pipe, size_t len,
366 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800367
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700368static inline void tcp_dec_quickack_mode(struct sock *sk,
369 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700371 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700372
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700373 if (icsk->icsk_ack.quick) {
374 if (pkts >= icsk->icsk_ack.quick) {
375 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700376 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700377 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700378 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700379 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 }
381}
382
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700383#define TCP_ECN_OK 1
384#define TCP_ECN_QUEUE_CWR 2
385#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000386#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700387
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000388enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 TCP_TW_SUCCESS = 0,
390 TCP_TW_RST = 1,
391 TCP_TW_ACK = 2,
392 TCP_TW_SYN = 3
393};
394
395
Joe Perches5c9f3022013-09-23 11:33:32 -0700396enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
397 struct sk_buff *skb,
398 const struct tcphdr *th);
399struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazet52452c52015-03-19 19:04:19 -0700400 struct request_sock *req, bool fastopen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700401int tcp_child_process(struct sock *parent, struct sock *child,
402 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400403void tcp_enter_loss(struct sock *sk);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800404void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
Joe Perches5c9f3022013-09-23 11:33:32 -0700405void tcp_clear_retrans(struct tcp_sock *tp);
406void tcp_update_metrics(struct sock *sk);
407void tcp_init_metrics(struct sock *sk);
408void tcp_metrics_init(void);
Soheil Hassas Yeganehd82bae12017-03-15 16:30:45 -0400409bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700410void tcp_disable_fack(struct tcp_sock *tp);
411void tcp_close(struct sock *sk, long timeout);
412void tcp_init_sock(struct sock *sk);
413unsigned int tcp_poll(struct file *file, struct socket *sock,
414 struct poll_table_struct *wait);
415int tcp_getsockopt(struct sock *sk, int level, int optname,
416 char __user *optval, int __user *optlen);
417int tcp_setsockopt(struct sock *sk, int level, int optname,
418 char __user *optval, unsigned int optlen);
419int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000420 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700421int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000422 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700423void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700424void tcp_syn_ack_timeout(const struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800425int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
426 int flags, int *addr_len);
Joe Perches5c9f3022013-09-23 11:33:32 -0700427void tcp_parse_options(const struct sk_buff *skb,
428 struct tcp_options_received *opt_rx,
429 int estab, struct tcp_fastopen_cookie *foc);
430const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432/*
433 * TCP v4 functions exported for the inet6 API
434 */
435
Joe Perches5c9f3022013-09-23 11:33:32 -0700436void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400437void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet9cf74902016-02-02 19:31:12 -0800438void tcp_req_err(struct sock *sk, u32 seq, bool abort);
Joe Perches5c9f3022013-09-23 11:33:32 -0700439int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700440struct sock *tcp_create_openreq_child(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -0700441 struct request_sock *req,
442 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100443void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Eric Dumazet0c271712015-09-29 07:42:48 -0700444struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700445 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700446 struct dst_entry *dst,
447 struct request_sock *req_unhash,
448 bool *own_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700449int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
450int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
451int tcp_connect(struct sock *sk);
Eric Dumazetb3d05142016-04-13 22:05:39 -0700452enum tcp_synack_type {
453 TCP_SYNACK_NORMAL,
454 TCP_SYNACK_FASTOPEN,
455 TCP_SYNACK_COOKIE,
456};
Eric Dumazet5d062de2015-09-25 07:39:19 -0700457struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
Joe Perches5c9f3022013-09-23 11:33:32 -0700458 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700459 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -0700460 enum tcp_synack_type synack_type);
Joe Perches5c9f3022013-09-23 11:33:32 -0700461int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000463void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000464int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000465void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467/* From syncookies.c */
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700468struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
469 struct request_sock *req,
470 struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700471int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
472 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700473struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400474#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200475
Eric Dumazet63262312014-03-19 21:02:21 -0700476/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200477 * This counter is used both as a hash input and partially encoded into
478 * the cookie value. A cookie is only validated further if the delta
479 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700480 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200481 * the counter advances immediately after a cookie is generated).
482 */
Eric Dumazet264ea102015-05-14 14:26:56 -0700483#define MAX_SYNCOOKIE_AGE 2
484#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
485#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
486
487/* syncookies: remember time of last synqueue overflow
488 * But do not dirty this field too often (once per second is enough)
Eric Dumazet3f684b42015-09-29 07:42:49 -0700489 * It is racy as we do not hold a lock, but race is very minor.
Eric Dumazet264ea102015-05-14 14:26:56 -0700490 */
Eric Dumazet3f684b42015-09-29 07:42:49 -0700491static inline void tcp_synq_overflow(const struct sock *sk)
Eric Dumazet264ea102015-05-14 14:26:56 -0700492{
493 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
494 unsigned long now = jiffies;
495
496 if (time_after(now, last_overflow + HZ))
497 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
498}
499
500/* syncookies: no recent synqueue overflow on this listening socket? */
501static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
502{
503 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
504
505 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
506}
Florian Westphal8c27bd72013-09-20 22:32:55 +0200507
508static inline u32 tcp_cookie_time(void)
509{
Eric Dumazet63262312014-03-19 21:02:21 -0700510 u64 val = get_jiffies_64();
511
Eric Dumazet264ea102015-05-14 14:26:56 -0700512 do_div(val, TCP_SYNCOOKIE_PERIOD);
Eric Dumazet63262312014-03-19 21:02:21 -0700513 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200514}
515
Joe Perches5c9f3022013-09-23 11:33:32 -0700516u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
517 u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700518__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
Joe Perches5c9f3022013-09-23 11:33:32 -0700519__u32 cookie_init_timestamp(struct request_sock *req);
Florian Westphalf1673382014-11-03 17:35:02 +0100520bool cookie_timestamp_decode(struct tcp_options_received *opt);
521bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100522 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700523
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800524/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700525int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
526 u32 cookie);
527struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100528
Joe Perches5c9f3022013-09-23 11:33:32 -0700529u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
530 const struct tcphdr *th, u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700531__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400532#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533/* tcp_output.c */
534
Neal Cardwell1b3878c2016-09-19 23:39:18 -0400535u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
536 int min_tso_segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700537void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
538 int nonagle);
539bool tcp_may_send_now(struct sock *sk);
Eric Dumazet10d3be52016-04-21 10:55:23 -0700540int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
541int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700542void tcp_retransmit_timer(struct sock *sk);
543void tcp_xmit_retransmit_queue(struct sock *);
544void tcp_simple_retransmit(struct sock *);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800545void tcp_enter_recovery(struct sock *sk, bool ece_ack);
Joe Perches5c9f3022013-09-23 11:33:32 -0700546int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300547int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
Joe Perches5c9f3022013-09-23 11:33:32 -0700549void tcp_send_probe0(struct sock *);
550void tcp_send_partial(struct sock *);
Eric Dumazete520af42015-05-06 14:26:25 -0700551int tcp_write_wakeup(struct sock *, int mib);
Joe Perches5c9f3022013-09-23 11:33:32 -0700552void tcp_send_fin(struct sock *sk);
553void tcp_send_active_reset(struct sock *sk, gfp_t priority);
554int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700555void tcp_push_one(struct sock *, unsigned int mss_now);
556void tcp_send_ack(struct sock *sk);
557void tcp_send_delayed_ack(struct sock *sk);
558void tcp_send_loss_probe(struct sock *sk);
559bool tcp_schedule_loss_probe(struct sock *sk);
Martin KaFai Laucfea5a62016-04-19 22:39:29 -0700560void tcp_skb_collapse_tstamp(struct sk_buff *skb,
561 const struct sk_buff *next_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
David S. Millera762a982005-07-05 15:18:51 -0700563/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700564void tcp_rearm_rto(struct sock *sk);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700565void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700566void tcp_reset(struct sock *sk);
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -0700567void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
Eric Dumazete3e17b72016-02-06 11:16:28 -0800568void tcp_fin(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700571void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700572static inline void tcp_clear_xmit_timers(struct sock *sk)
573{
574 inet_csk_clear_xmit_timers(sk);
575}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Joe Perches5c9f3022013-09-23 11:33:32 -0700577unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
578unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000579
580/* Bound MSS / TSO packet size with the half of the window */
581static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
582{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700583 int cutoff;
584
585 /* When peer uses tiny windows, there is no use in packetizing
586 * to sub-MSS pieces for the sake of SWS or making sure there
587 * are enough packets in the pipe for fast recovery.
588 *
589 * On the other hand, for extremely large MSS devices, handling
590 * smaller than MSS windows in this way does make sense.
591 */
Seymour, Shane M2631b792016-06-28 23:06:48 +0000592 if (tp->max_window > TCP_MSS_DEFAULT)
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700593 cutoff = (tp->max_window >> 1);
594 else
595 cutoff = tp->max_window;
596
597 if (cutoff && pktsize > cutoff)
598 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000599 else
600 return pktsize;
601}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300603/* tcp.c */
Eric Dumazet0df48c22015-04-28 15:28:17 -0700604void tcp_get_info(struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606/* Read 'sendfile()'-style from a TCP socket */
Joe Perches5c9f3022013-09-23 11:33:32 -0700607int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
608 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Joe Perches5c9f3022013-09-23 11:33:32 -0700610void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Joe Perches5c9f3022013-09-23 11:33:32 -0700612int tcp_mtu_to_mss(struct sock *sk, int pmtu);
613int tcp_mss_to_mtu(struct sock *sk, int mss);
614void tcp_mtup_init(struct sock *sk);
615void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800616
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000617static inline void tcp_bound_rto(const struct sock *sk)
618{
619 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
620 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
621}
622
623static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
624{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800625 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000626}
627
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800628static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
631 ntohl(TCP_FLAG_ACK) |
632 snd_wnd);
633}
634
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800635static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
637 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
638}
639
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700640static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700642 struct tcp_sock *tp = tcp_sk(sk);
643
Yaogong Wang9f5afea2016-09-07 14:49:28 -0700644 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 tp->rcv_wnd &&
646 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
647 !tp->urg_data)
648 tcp_fast_path_on(tp);
649}
650
Satoru SATOH0c266892009-05-04 11:11:01 -0700651/* Compute the actual rto_min value */
652static inline u32 tcp_rto_min(struct sock *sk)
653{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400654 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700655 u32 rto_min = TCP_RTO_MIN;
656
657 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
658 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
659 return rto_min;
660}
661
Eric Dumazet740b0f12014-02-26 14:02:48 -0800662static inline u32 tcp_rto_min_us(struct sock *sk)
663{
664 return jiffies_to_usecs(tcp_rto_min(sk));
665}
666
Daniel Borkmann81164412015-01-05 23:57:48 +0100667static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
668{
669 return dst_metric_locked(dst, RTAX_CC_ALGO);
670}
671
Yuchung Chengf6722582015-10-16 21:57:42 -0700672/* Minimum RTT in usec. ~0 means not available. */
673static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
674{
Neal Cardwell64033892016-09-19 23:39:10 -0400675 return minmax_get(&tp->rtt_min);
Yuchung Chengf6722582015-10-16 21:57:42 -0700676}
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678/* Compute the actual receive window we are currently advertising.
679 * Rcv_nxt can be after the window if our peer push more data
680 * than the offered window.
681 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800682static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
685
686 if (win < 0)
687 win = 0;
688 return (u32) win;
689}
690
691/* Choose a new window, without checks for shrinking, and without
692 * scaling applied to the result. The caller does these things
693 * if necessary. This is a "raw" window selection.
694 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700695u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Pavel Emelyanovee995282012-04-19 03:40:39 +0000697void tcp_send_window_probe(struct sock *sk);
698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699/* TCP timestamps are only 32-bits, this causes a slight
700 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800701 * of jiffies in the buffer control blocks below. We decided
702 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 * casts with the following macro.
704 */
705#define tcp_time_stamp ((__u32)(jiffies))
706
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700707static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
708{
709 return skb->skb_mstamp.stamp_jiffies;
710}
711
712
Changli Gaoa3433f32010-06-12 14:01:43 +0000713#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
714
715#define TCPHDR_FIN 0x01
716#define TCPHDR_SYN 0x02
717#define TCPHDR_RST 0x04
718#define TCPHDR_PSH 0x08
719#define TCPHDR_ACK 0x10
720#define TCPHDR_URG 0x20
721#define TCPHDR_ECE 0x40
722#define TCPHDR_CWR 0x80
723
Daniel Borkmann49213552015-05-19 21:04:22 +0200724#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
725
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800726/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700727 * TCP per-packet control information to the transmission code.
728 * We also store the host-order sequence numbers in here too.
729 * This is 44 bytes if IPV6 is enabled.
730 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
732struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 __u32 seq; /* Starting sequence number */
734 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700735 union {
736 /* Note : tcp_tw_isn is used in input path only
737 * (isn chosen by tcp_timewait_state_process())
738 *
Eric Dumazetf69ad292015-06-11 09:15:18 -0700739 * tcp_gso_segs/size are used in write queue only,
740 * cf tcp_skb_pcount()/tcp_skb_mss()
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700741 */
742 __u32 tcp_tw_isn;
Eric Dumazetf69ad292015-06-11 09:15:18 -0700743 struct {
744 u16 tcp_gso_segs;
745 u16 tcp_gso_size;
746 };
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700747 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400748 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 __u8 sacked; /* State flags for SACK/FACK. */
751#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
752#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
753#define TCPCB_LOST 0x04 /* SKB is lost */
754#define TCPCB_TAGBITS 0x07 /* All tag bits */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400755#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400757#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
758 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000760 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
Soheil Hassas Yeganeh6b084922016-04-02 23:08:08 -0400761 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700762 eor:1, /* Is skb MSG_EOR marked? */
763 unused:6;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700765 union {
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700766 struct {
Yuchung Chengb9f64822016-09-19 23:39:14 -0400767 /* There is space for up to 24 bytes */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400768 __u32 in_flight:30,/* Bytes in flight at transmit */
769 is_app_limited:1, /* cwnd not fully used? */
770 unused:1;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400771 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
772 __u32 delivered;
773 /* start of send pipeline phase */
774 struct skb_mstamp first_tx_mstamp;
775 /* when we reached the "delivered" count */
776 struct skb_mstamp delivered_mstamp;
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700777 } tx; /* only used for outgoing skbs */
778 union {
779 struct inet_skb_parm h4;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700780#if IS_ENABLED(CONFIG_IPV6)
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700781 struct inet6_skb_parm h6;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700782#endif
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700783 } header; /* For incoming skbs */
784 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785};
786
787#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
788
Eric Dumazet870c3152014-10-17 09:17:20 -0700789
Eric Dumazet815afe12014-10-18 08:34:37 -0700790#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700791/* This is the variant of inet6_iif() that must be used by TCP,
792 * as TCP moves IP6CB into a different location in skb->cb[]
793 */
794static inline int tcp_v6_iif(const struct sk_buff *skb)
795{
David Aherna04a4802016-10-16 20:02:52 -0700796 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
David Ahern74b20582016-05-10 11:19:50 -0700797
798 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
Eric Dumazet870c3152014-10-17 09:17:20 -0700799}
Eric Dumazet815afe12014-10-18 08:34:37 -0700800#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700801
David Aherna04a4802016-10-16 20:02:52 -0700802/* TCP_SKB_CB reference means this can not be used from early demux */
803static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
804{
805#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
806 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
David Ahernda967862016-11-02 12:08:25 -0700807 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
David Aherna04a4802016-10-16 20:02:52 -0700808 return true;
809#endif
810 return false;
811}
812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813/* Due to TSO, an SKB can be composed of multiple actual
814 * packets. To keep these tracked properly, we use this.
815 */
816static inline int tcp_skb_pcount(const struct sk_buff *skb)
817{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700818 return TCP_SKB_CB(skb)->tcp_gso_segs;
819}
820
821static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
822{
823 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
824}
825
826static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
827{
828 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829}
830
Eric Dumazetf69ad292015-06-11 09:15:18 -0700831/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832static inline int tcp_skb_mss(const struct sk_buff *skb)
833{
Eric Dumazetf69ad292015-06-11 09:15:18 -0700834 return TCP_SKB_CB(skb)->tcp_gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
836
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700837static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
838{
839 return likely(!TCP_SKB_CB(skb)->eor);
840}
841
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700842/* Events passed to congestion control interface */
843enum tcp_ca_event {
844 CA_EVENT_TX_START, /* first transmit when no packets in flight */
845 CA_EVENT_CWND_RESTART, /* congestion window restart */
846 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700847 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200848 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
849 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
850 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
851 CA_EVENT_NON_DELAYED_ACK,
Florian Westphal7354c8c2014-09-26 22:37:34 +0200852};
853
Florian Westphal98900922014-09-26 22:37:35 +0200854/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200855enum tcp_ca_ack_event_flags {
Florian Westphal98900922014-09-26 22:37:35 +0200856 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
857 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
858 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700859};
860
861/*
862 * Interface for adding new TCP congestion control handlers
863 */
864#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800865#define TCP_CA_MAX 128
866#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
867
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100868#define TCP_CA_UNSPEC 0
869
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200870/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700871#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200872/* Requires ECN/ECT set on all packets */
873#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700874
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700875union tcp_cc_info;
876
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700877struct ack_sample {
878 u32 pkts_acked;
879 s32 rtt_us;
Lawrence Brakmo6f094b92016-06-08 21:16:44 -0700880 u32 in_flight;
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700881};
882
Yuchung Chengb9f64822016-09-19 23:39:14 -0400883/* A rate sample measures the number of (original/retransmitted) data
884 * packets delivered "delivered" over an interval of time "interval_us".
885 * The tcp_rate.c code fills in the rate sample, and congestion
886 * control modules that define a cong_control function to run at the end
887 * of ACK processing can optionally chose to consult this sample when
888 * setting cwnd and pacing rate.
889 * A sample is invalid if "delivered" or "interval_us" is negative.
890 */
891struct rate_sample {
892 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
893 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
894 s32 delivered; /* number of packets delivered over interval */
895 long interval_us; /* time for tp->delivered to incr "delivered" */
896 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
897 int losses; /* number of packets marked lost upon ACK */
898 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
899 u32 prior_in_flight; /* in flight before this ACK */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400900 bool is_app_limited; /* is sample from packet with bubble in pipe? */
Yuchung Chengb9f64822016-09-19 23:39:14 -0400901 bool is_retrans; /* is sample from retransmission? */
902};
903
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700904struct tcp_congestion_ops {
905 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100906 u32 key;
907 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700908
909 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300910 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700911 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300912 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700913
914 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300915 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700916 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700917 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700918 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300919 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700920 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300921 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +0200922 /* call when ack arrives (optional) */
923 void (*in_ack_event)(struct sock *sk, u32 flags);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700924 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300925 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700926 /* hook for packet ack accounting (optional) */
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700927 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
Neal Cardwelled6e7262016-09-19 23:39:17 -0400928 /* suggest number of segments for each skb to transmit (optional) */
929 u32 (*tso_segs_goal)(struct sock *sk);
Yuchung Cheng77bfc172016-09-19 23:39:20 -0400930 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
931 u32 (*sndbuf_expand)(struct sock *sk);
Yuchung Chengc0402762016-09-19 23:39:21 -0400932 /* call when packets are delivered to update cwnd and pacing rate,
933 * after all the ca_state processing. (optional)
934 */
935 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300936 /* get info for inet_diag (optional) */
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700937 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
938 union tcp_cc_info *info);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700939
940 char name[TCP_CA_NAME_MAX];
941 struct module *owner;
942};
943
Joe Perches5c9f3022013-09-23 11:33:32 -0700944int tcp_register_congestion_control(struct tcp_congestion_ops *type);
945void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700946
Florian Westphal55d86942014-09-26 22:37:32 +0200947void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700948void tcp_init_congestion_control(struct sock *sk);
949void tcp_cleanup_congestion_control(struct sock *sk);
950int tcp_set_default_congestion_control(const char *name);
951void tcp_get_default_congestion_control(char *name);
952void tcp_get_available_congestion_control(char *buf, size_t len);
953void tcp_get_allowed_congestion_control(char *buf, size_t len);
954int tcp_set_allowed_congestion_control(char *allowed);
955int tcp_set_congestion_control(struct sock *sk, const char *name);
Neal Cardwelle73ebb082015-01-28 20:01:35 -0500956u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
957void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700958
Joe Perches5c9f3022013-09-23 11:33:32 -0700959u32 tcp_reno_ssthresh(struct sock *sk);
Florian Westphale9799182016-11-21 14:18:38 +0100960u32 tcp_reno_undo_cwnd(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -0700961void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -0700962extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700963
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100964struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
Daniel Borkmannc3a8d942015-08-31 15:58:47 +0200965u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
Daniel Borkmannea697632015-01-05 23:57:47 +0100966#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100967char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +0100968#else
969static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
970{
971 return NULL;
972}
973#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100974
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200975static inline bool tcp_ca_needs_ecn(const struct sock *sk)
976{
977 const struct inet_connection_sock *icsk = inet_csk(sk);
978
979 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
980}
981
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300982static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700983{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300984 struct inet_connection_sock *icsk = inet_csk(sk);
985
986 if (icsk->icsk_ca_ops->set_state)
987 icsk->icsk_ca_ops->set_state(sk, ca_state);
988 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700989}
990
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300991static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700992{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300993 const struct inet_connection_sock *icsk = inet_csk(sk);
994
995 if (icsk->icsk_ca_ops->cwnd_event)
996 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700997}
998
Yuchung Chengb9f64822016-09-19 23:39:14 -0400999/* From tcp_rate.c */
1000void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1001void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1002 struct rate_sample *rs);
1003void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1004 struct skb_mstamp *now, struct rate_sample *rs);
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001005void tcp_rate_check_app_limited(struct sock *sk);
Yuchung Chengb9f64822016-09-19 23:39:14 -04001006
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001007/* These functions determine how the current flow behaves in respect of SACK
1008 * handling. SACK is negotiated with the peer, and therefore it can vary
1009 * between different flows.
1010 *
1011 * tcp_is_sack - SACK enabled
1012 * tcp_is_reno - No SACK
1013 * tcp_is_fack - FACK enabled, implies SACK enabled
1014 */
1015static inline int tcp_is_sack(const struct tcp_sock *tp)
1016{
1017 return tp->rx_opt.sack_ok;
1018}
1019
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001020static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001021{
1022 return !tcp_is_sack(tp);
1023}
1024
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001025static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001026{
Vijay Subramanianab562222011-12-20 13:23:24 +00001027 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001028}
1029
1030static inline void tcp_enable_fack(struct tcp_sock *tp)
1031{
Vijay Subramanianab562222011-12-20 13:23:24 +00001032 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001033}
1034
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001035static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1036{
1037 return tp->sacked_out + tp->lost_out;
1038}
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040/* This determines how many packets are "in the network" to the best
1041 * of our knowledge. In many cases it is conservative, but where
1042 * detailed information is available from the receiver (via SACK
1043 * blocks etc.) we can make more aggressive calculations.
1044 *
1045 * Use this for decisions involving congestion control, use just
1046 * tp->packets_out to determine if the send queue is empty or not.
1047 *
1048 * Read this equation as:
1049 *
1050 * "Packets sent once on transmission queue" MINUS
1051 * "Packets left network, but not honestly ACKed yet" PLUS
1052 * "Packets fast retransmitted"
1053 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001054static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001056 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057}
1058
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001059#define TCP_INFINITE_SSTHRESH 0x7fffffff
1060
Yuchung Cheng071d5082015-07-09 13:16:29 -07001061static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1062{
Yuchung Cheng76174002015-07-09 13:16:30 -07001063 return tp->snd_cwnd < tp->snd_ssthresh;
Yuchung Cheng071d5082015-07-09 13:16:29 -07001064}
1065
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001066static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1067{
1068 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1069}
1070
Yuchung Cheng684bad12012-09-02 17:38:04 +00001071static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1072{
1073 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1074 (1 << inet_csk(sk)->icsk_ca_state);
1075}
1076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +00001078 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 * ssthresh.
1080 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001081static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001083 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001084
Yuchung Cheng684bad12012-09-02 17:38:04 +00001085 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 return tp->snd_ssthresh;
1087 else
1088 return max(tp->snd_ssthresh,
1089 ((tp->snd_cwnd >> 1) +
1090 (tp->snd_cwnd >> 2)));
1091}
1092
Ilpo Järvinenb9c45952007-07-27 16:36:17 +03001093/* Use define here intentionally to get WARN_ON location shown at the caller */
1094#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Christoph Paasch5ee2c942014-07-14 16:58:32 +02001096void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001097__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098
Neal Cardwell6b5a5c02011-11-21 17:15:14 +00001099/* The maximum number of MSS of available cwnd for which TSO defers
1100 * sending if not using sysctl_tcp_tso_win_divisor.
1101 */
1102static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1103{
1104 return 3;
1105}
1106
Ilpo Järvinen90840de2007-12-31 04:48:41 -08001107/* Returns end sequence number of the receiver's advertised window */
1108static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1109{
1110 return tp->snd_una + tp->snd_wnd;
1111}
Eric Dumazete114a712014-04-30 11:58:13 -07001112
1113/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1114 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001115 * it was fully used previously. And that's exactly what we do in
1116 * congestion avoidance mode. But in slow start we allow cwnd to grow
1117 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001118 * Example :
1119 * cwnd is 10 (IW10), but application sends 9 frames.
1120 * We allow cwnd to reach 18 when all frames are ACKed.
1121 * This check is safe because it's as aggressive as slow start which already
1122 * risks 100% overshoot. The advantage is that we discourage application to
1123 * either send more filler packets or data to artificially blow up the cwnd
1124 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001125 */
Eric Dumazet24901552014-05-02 21:18:05 -07001126static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001127{
1128 const struct tcp_sock *tp = tcp_sk(sk);
1129
Neal Cardwellca8a2262014-05-22 10:41:08 -04001130 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
Yuchung Cheng071d5082015-07-09 13:16:29 -07001131 if (tcp_in_slow_start(tp))
Neal Cardwellca8a2262014-05-22 10:41:08 -04001132 return tp->snd_cwnd < 2 * tp->max_packets_out;
1133
1134 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001135}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001136
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001137/* Something is really bad, we could not queue an additional packet,
1138 * because qdisc is full or receiver sent a 0 window.
1139 * We do not want to add fuel to the fire, or abort too early,
1140 * so make sure the timer we arm now is at least 200ms in the future,
1141 * regardless of current icsk_rto value (as it could be ~2ms)
1142 */
1143static inline unsigned long tcp_probe0_base(const struct sock *sk)
1144{
1145 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1146}
1147
1148/* Variant of inet_csk_rto_backoff() used for zero window probes */
1149static inline unsigned long tcp_probe0_when(const struct sock *sk,
1150 unsigned long max_when)
1151{
1152 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1153
1154 return (unsigned long)min_t(u64, when, max_when);
1155}
1156
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001157static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001159 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001160 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001161 tcp_probe0_base(sk), TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162}
1163
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001164static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165{
1166 tp->snd_wl1 = seq;
1167}
1168
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001169static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
1171 tp->snd_wl1 = seq;
1172}
1173
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174/*
1175 * Calculate(/check) TCP checksum
1176 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001177static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1178 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179{
1180 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1181}
1182
Al Virob51655b2006-11-14 21:40:42 -08001183static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
Herbert Xufb286bb2005-11-10 13:01:24 -08001185 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001188static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Herbert Xu60476372007-04-09 11:59:39 -07001190 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 __tcp_checksum_complete(skb);
1192}
1193
1194/* Prequeue for VJ style copy to user, combined with checksumming. */
1195
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001196static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
1198 tp->ucopy.task = NULL;
1199 tp->ucopy.len = 0;
1200 tp->ucopy.memory = 0;
1201 skb_queue_head_init(&tp->ucopy.prequeue);
1202}
1203
Joe Perches5c9f3022013-09-23 11:33:32 -07001204bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc9c33212016-08-27 07:37:54 -07001205bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
Eric Dumazetac6e7802016-11-10 13:12:35 -08001206int tcp_filter(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
1208#undef STATE_TRACE
1209
1210#ifdef STATE_TRACE
1211static const char *statename[]={
1212 "Unused","Established","Syn Sent","Syn Recv",
1213 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1214 "Close Wait","Last ACK","Listen","Closing"
1215};
1216#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001217void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Joe Perches5c9f3022013-09-23 11:33:32 -07001219void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09001221int tcp_abort(struct sock *sk, int err);
1222
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001223static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224{
1225 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 rx_opt->num_sacks = 0;
1227}
1228
Joe Perches5c9f3022013-09-23 11:33:32 -07001229u32 tcp_default_init_rwnd(u32 mss);
Eric Dumazet6f021c62015-08-21 12:30:00 -07001230void tcp_cwnd_restart(struct sock *sk, s32 delta);
1231
1232static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1233{
1234 struct tcp_sock *tp = tcp_sk(sk);
1235 s32 delta;
1236
1237 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1238 return;
1239 delta = tcp_time_stamp - tp->lsndtime;
1240 if (delta > inet_csk(sk)->icsk_rto)
1241 tcp_cwnd_restart(sk, delta);
1242}
Yuchung Cheng85f16522013-06-11 15:35:32 -07001243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001245void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1246 __u32 *window_clamp, int wscale_ok,
1247 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
1249static inline int tcp_win_from_space(int space)
1250{
1251 return sysctl_tcp_adv_win_scale<=0 ?
1252 (space>>(-sysctl_tcp_adv_win_scale)) :
1253 space - (space>>sysctl_tcp_adv_win_scale);
1254}
1255
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001256/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257static inline int tcp_space(const struct sock *sk)
1258{
1259 return tcp_win_from_space(sk->sk_rcvbuf -
1260 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001261}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
1263static inline int tcp_full_space(const struct sock *sk)
1264{
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001265 return tcp_win_from_space(sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266}
1267
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001268extern void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -07001269 const struct sock *sk_listener,
1270 const struct dst_entry *dst);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001271
Joe Perches5c9f3022013-09-23 11:33:32 -07001272void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1275{
Nikolay Borisovb840d152016-01-07 16:38:45 +02001276 struct net *net = sock_net((struct sock *)tp);
1277
1278 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
1280
1281static inline int keepalive_time_when(const struct tcp_sock *tp)
1282{
Nikolay Borisov13b287e2016-01-07 16:38:43 +02001283 struct net *net = sock_net((struct sock *)tp);
1284
1285 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286}
1287
Eric Dumazetdf19a622009-08-28 23:48:54 -07001288static inline int keepalive_probes(const struct tcp_sock *tp)
1289{
Nikolay Borisov9bd68612016-01-07 16:38:44 +02001290 struct net *net = sock_net((struct sock *)tp);
1291
1292 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
Eric Dumazetdf19a622009-08-28 23:48:54 -07001293}
1294
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001295static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1296{
1297 const struct inet_connection_sock *icsk = &tp->inet_conn;
1298
1299 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1300 tcp_time_stamp - tp->rcv_tstamp);
1301}
1302
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001303static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02001305 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001306 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001308 if (fin_timeout < (rto << 2) - (rto >> 1))
1309 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311 return fin_timeout;
1312}
1313
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001314static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1315 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001317 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001318 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001319 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001320 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001321 /*
1322 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1323 * then following tcp messages have valid values. Ignore 0 value,
1324 * or else 'negative' tsval might forbid us to accept their packets.
1325 */
1326 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001327 return true;
1328 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001329}
1330
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001331static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1332 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001333{
1334 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001335 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 /* RST segments are not recommended to carry timestamp,
1338 and, if they do, it is recommended to ignore PAWS because
1339 "their cleanup function should take precedence over timestamps."
1340 Certainly, it is mistake. It is necessary to understand the reasons
1341 of this constraint to relax it: if peer reboots, clock may go
1342 out-of-sync and half-open connections will not be reset.
1343 Actually, the problem would be not existing if all
1344 the implementations followed draft about maintaining clock
1345 via reboots. Linux-2.2 DOES NOT!
1346
1347 However, we can relax time bounds for RST segments to MSL.
1348 */
James Morris9d729f72007-03-04 16:12:44 -08001349 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001350 return false;
1351 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352}
1353
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001354bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1355 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001356
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001357static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358{
1359 /* See RFC 2012 */
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001360 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1361 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1362 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1363 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364}
1365
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001366/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001367static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001368{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001369 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001370}
1371
1372static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1373{
1374 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001375 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001376}
1377
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001378union tcp_md5_addr {
1379 struct in_addr a4;
1380#if IS_ENABLED(CONFIG_IPV6)
1381 struct in6_addr a6;
1382#endif
1383};
1384
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001385/* - key database */
1386struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001387 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001388 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001389 u8 family; /* AF_INET or AF_INET6 */
1390 union tcp_md5_addr addr;
1391 u8 key[TCP_MD5SIG_MAXKEYLEN];
1392 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001393};
1394
1395/* - sock block */
1396struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001397 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001398 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001399};
1400
1401/* - pseudo header */
1402struct tcp4_pseudohdr {
1403 __be32 saddr;
1404 __be32 daddr;
1405 __u8 pad;
1406 __u8 protocol;
1407 __be16 len;
1408};
1409
1410struct tcp6_pseudohdr {
1411 struct in6_addr saddr;
1412 struct in6_addr daddr;
1413 __be32 len;
1414 __be32 protocol; /* including padding */
1415};
1416
1417union tcp_md5sum_block {
1418 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001419#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001420 struct tcp6_pseudohdr ip6;
1421#endif
1422};
1423
1424/* - pool: digest algorithm, hash description and scratch buffer */
1425struct tcp_md5sig_pool {
Herbert Xucf80e0e2016-01-24 21:20:23 +08001426 struct ahash_request *md5_req;
Eric Dumazet19689e32016-06-27 18:51:53 +02001427 void *scratch;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001428};
1429
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001430/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001431int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1432 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001433int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1434 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1435int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1436 int family);
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001437struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001438 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001439
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001440#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001441struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -07001442 const union tcp_md5_addr *addr,
1443 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001444#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001445#else
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001446static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001447 const union tcp_md5_addr *addr,
1448 int family)
1449{
1450 return NULL;
1451}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001452#define tcp_twsk_md5_key(twsk) NULL
1453#endif
1454
Joe Perches5c9f3022013-09-23 11:33:32 -07001455bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001456
Joe Perches5c9f3022013-09-23 11:33:32 -07001457struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001458static inline void tcp_put_md5sig_pool(void)
1459{
1460 local_bh_enable();
1461}
Eric Dumazet35790c02010-05-16 00:34:04 -07001462
Joe Perches5c9f3022013-09-23 11:33:32 -07001463int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1464 unsigned int header_len);
1465int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1466 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001467
Jerry Chu10467162012-08-31 12:29:11 +00001468/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001469void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1470 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1471 unsigned long *last_syn_loss);
1472void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -07001473 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1474 u16 try_exp);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001475struct tcp_fastopen_request {
1476 /* Fast Open cookie. Size 0 means a cookie request */
1477 struct tcp_fastopen_cookie cookie;
1478 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001479 size_t size;
1480 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001481};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001482void tcp_free_fastopen_req(struct tcp_sock *tp);
1483
Jerry Chu10467162012-08-31 12:29:11 +00001484extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1485int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Eric Dumazet61d2bca2016-02-01 21:03:07 -08001486void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
Eric Dumazet7c85af82015-09-24 17:16:05 -07001487struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1488 struct request_sock *req,
1489 struct tcp_fastopen_cookie *foc,
1490 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001491void tcp_fastopen_init_key_once(bool publish);
Wei Wang065263f2017-01-23 10:59:20 -08001492bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1493 struct tcp_fastopen_cookie *cookie);
Wei Wang19f6d3f2017-01-23 10:59:22 -08001494bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
Jerry Chu10467162012-08-31 12:29:11 +00001495#define TCP_FASTOPEN_KEY_LENGTH 16
1496
1497/* Fastopen key context */
1498struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001499 struct crypto_cipher *tfm;
1500 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1501 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001502};
1503
Francis Yan05b055e2016-11-27 23:07:13 -08001504/* Latencies incurred by various limits for a sender. They are
1505 * chronograph-like stats that are mutually exclusive.
1506 */
1507enum tcp_chrono {
1508 TCP_CHRONO_UNSPEC,
1509 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1510 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1511 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1512 __TCP_CHRONO_MAX,
1513};
1514
1515void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1516void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1517
David S. Millerfe067e82007-03-07 12:12:44 -08001518/* write queue abstraction */
1519static inline void tcp_write_queue_purge(struct sock *sk)
1520{
1521 struct sk_buff *skb;
1522
Francis Yan0f872302016-11-27 23:07:14 -08001523 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
David S. Millerfe067e82007-03-07 12:12:44 -08001524 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001525 sk_wmem_free_skb(sk, skb);
1526 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001527 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001528}
1529
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001530static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001531{
David S. Millercd07a8e2008-09-23 00:50:13 -07001532 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001533}
1534
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001535static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001536{
David S. Millercd07a8e2008-09-23 00:50:13 -07001537 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001538}
1539
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001540static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1541 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001542{
David S. Millercd07a8e2008-09-23 00:50:13 -07001543 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001544}
1545
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001546static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1547 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001548{
1549 return skb_queue_prev(&sk->sk_write_queue, skb);
1550}
1551
David S. Millerfe067e82007-03-07 12:12:44 -08001552#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001553 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001554
1555#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001556 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001557
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001558#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001559 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001560
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001561static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001562{
1563 return sk->sk_send_head;
1564}
1565
David S. Millercd07a8e2008-09-23 00:50:13 -07001566static inline bool tcp_skb_is_last(const struct sock *sk,
1567 const struct sk_buff *skb)
1568{
1569 return skb_queue_is_last(&sk->sk_write_queue, skb);
1570}
1571
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001572static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001573{
David S. Millercd07a8e2008-09-23 00:50:13 -07001574 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001575 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001576 else
1577 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001578}
1579
1580static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1581{
Francis Yan0f872302016-11-27 23:07:14 -08001582 if (sk->sk_send_head == skb_unlinked) {
David S. Millerfe067e82007-03-07 12:12:44 -08001583 sk->sk_send_head = NULL;
Francis Yan0f872302016-11-27 23:07:14 -08001584 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1585 }
Eric Dumazetbb1fcec2016-08-17 05:56:26 -07001586 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1587 tcp_sk(sk)->highest_sack = NULL;
David S. Millerfe067e82007-03-07 12:12:44 -08001588}
1589
1590static inline void tcp_init_send_head(struct sock *sk)
1591{
1592 sk->sk_send_head = NULL;
1593}
1594
1595static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1596{
1597 __skb_queue_tail(&sk->sk_write_queue, skb);
1598}
1599
1600static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1601{
1602 __tcp_add_write_queue_tail(sk, skb);
1603
1604 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001605 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001606 sk->sk_send_head = skb;
Francis Yan0f872302016-11-27 23:07:14 -08001607 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001608
1609 if (tcp_sk(sk)->highest_sack == NULL)
1610 tcp_sk(sk)->highest_sack = skb;
1611 }
David S. Millerfe067e82007-03-07 12:12:44 -08001612}
1613
1614static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1615{
1616 __skb_queue_head(&sk->sk_write_queue, skb);
1617}
1618
1619/* Insert buff after skb on the write queue of sk. */
1620static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1621 struct sk_buff *buff,
1622 struct sock *sk)
1623{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001624 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001625}
1626
David S. Miller43f59c82008-09-21 21:28:51 -07001627/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001628static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1629 struct sk_buff *skb,
1630 struct sock *sk)
1631{
David S. Miller43f59c82008-09-21 21:28:51 -07001632 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001633
1634 if (sk->sk_send_head == skb)
1635 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001636}
1637
1638static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1639{
1640 __skb_unlink(skb, &sk->sk_write_queue);
1641}
1642
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001643static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001644{
1645 return skb_queue_empty(&sk->sk_write_queue);
1646}
1647
Krishna Kumar12d50c42009-12-08 22:26:13 +00001648static inline void tcp_push_pending_frames(struct sock *sk)
1649{
1650 if (tcp_send_head(sk)) {
1651 struct tcp_sock *tp = tcp_sk(sk);
1652
1653 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1654 }
1655}
1656
Neal Cardwellecb97192012-02-27 17:52:52 -05001657/* Start sequence of the skb just after the highest skb with SACKed
1658 * bit, valid only if sacked_out > 0 or when the caller has ensured
1659 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001660 */
1661static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1662{
1663 if (!tp->sacked_out)
1664 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001665
1666 if (tp->highest_sack == NULL)
1667 return tp->snd_nxt;
1668
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001669 return TCP_SKB_CB(tp->highest_sack)->seq;
1670}
1671
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001672static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1673{
1674 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1675 tcp_write_queue_next(sk, skb);
1676}
1677
1678static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1679{
1680 return tcp_sk(sk)->highest_sack;
1681}
1682
1683static inline void tcp_highest_sack_reset(struct sock *sk)
1684{
1685 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1686}
1687
1688/* Called when old skb is about to be deleted (to be combined with new skb) */
1689static inline void tcp_highest_sack_combine(struct sock *sk,
1690 struct sk_buff *old,
1691 struct sk_buff *new)
1692{
1693 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1694 tcp_sk(sk)->highest_sack = new;
1695}
1696
Florian Westphalb1f0a0e2015-12-21 21:29:24 +01001697/* This helper checks if socket has IP_TRANSPARENT set */
1698static inline bool inet_sk_transparent(const struct sock *sk)
1699{
1700 switch (sk->sk_state) {
1701 case TCP_TIME_WAIT:
1702 return inet_twsk(sk)->tw_transparent;
1703 case TCP_NEW_SYN_RECV:
1704 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1705 }
1706 return inet_sk(sk)->transparent;
1707}
1708
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001709/* Determines whether this is a thin stream (which may suffer from
1710 * increased latency). Used to trigger latency-reducing mechanisms.
1711 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001712static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001713{
1714 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1715}
1716
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717/* /proc */
1718enum tcp_seq_states {
1719 TCP_SEQ_STATE_LISTENING,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721};
1722
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001723int tcp_seq_open(struct inode *inode, struct file *file);
1724
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001726 char *name;
1727 sa_family_t family;
1728 const struct file_operations *seq_fops;
1729 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730};
1731
1732struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001733 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 sa_family_t family;
1735 enum tcp_seq_states state;
1736 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001737 int bucket, offset, sbucket, num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001738 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739};
1740
Joe Perches5c9f3022013-09-23 11:33:32 -07001741int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1742void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001744extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001745extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001746
Joe Perches5c9f3022013-09-23 11:33:32 -07001747void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001748
Eric Dumazet28be6e02013-10-18 10:36:17 -07001749struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001750 netdev_features_t features);
1751struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1752int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001753
Joe Perches5c9f3022013-09-23 11:33:32 -07001754void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001755
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001756static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1757{
Nikolay Borisov4979f2d2016-02-03 09:46:57 +02001758 struct net *net = sock_net((struct sock *)tp);
1759 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001760}
1761
1762static inline bool tcp_stream_memory_free(const struct sock *sk)
1763{
1764 const struct tcp_sock *tp = tcp_sk(sk);
1765 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1766
1767 return notsent_bytes < tcp_notsent_lowat(tp);
1768}
1769
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001770#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001771int tcp4_proc_init(void);
1772void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001773#endif
1774
Eric Dumazetea3bea32015-09-25 07:39:23 -07001775int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001776int tcp_conn_request(struct request_sock_ops *rsk_ops,
1777 const struct tcp_request_sock_ops *af_ops,
1778 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001779
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001780/* TCP af-specific functions */
1781struct tcp_sock_af_ops {
1782#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001783 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001784 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001785 int (*calc_md5_hash)(char *location,
1786 const struct tcp_md5sig_key *md5,
1787 const struct sock *sk,
1788 const struct sk_buff *skb);
1789 int (*md5_parse)(struct sock *sk,
1790 char __user *optval,
1791 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001792#endif
1793};
1794
1795struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001796 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001797#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001798 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001799 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001800 int (*calc_md5_hash) (char *location,
1801 const struct tcp_md5sig_key *md5,
1802 const struct sock *sk,
1803 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001804#endif
Eric Dumazetb40cf182015-09-25 07:39:08 -07001805 void (*init_req)(struct request_sock *req,
1806 const struct sock *sk_listener,
Octavian Purdila16bea702014-06-25 17:09:53 +03001807 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001808#ifdef CONFIG_SYN_COOKIES
Eric Dumazet3f684b42015-09-29 07:42:49 -07001809 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001810 __u16 *mss);
1811#endif
Eric Dumazetf9646292015-09-29 07:42:50 -07001812 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
Octavian Purdilad94e0412014-06-25 17:09:55 +03001813 const struct request_sock *req,
1814 bool *strict);
Alexey Kodaneva30aad52017-03-09 13:53:55 +03001815 __u32 (*init_seq_tsoff)(const struct sk_buff *skb, u32 *tsoff);
Eric Dumazet0f935db2015-09-25 07:39:21 -07001816 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001817 struct flowi *fl, struct request_sock *req,
Eric Dumazetdc6ef6b2015-10-16 13:00:01 -07001818 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -07001819 enum tcp_synack_type synack_type);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001820};
1821
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001822#ifdef CONFIG_SYN_COOKIES
1823static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001824 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001825 __u16 *mss)
1826{
Eric Dumazet3f684b42015-09-29 07:42:49 -07001827 tcp_synq_overflow(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001828 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
Eric Dumazet3f684b42015-09-29 07:42:49 -07001829 return ops->cookie_init_seq(skb, mss);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001830}
1831#else
1832static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001833 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001834 __u16 *mss)
1835{
1836 return 0;
1837}
1838#endif
1839
Joe Perches5c9f3022013-09-23 11:33:32 -07001840int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001841
Joe Perches5c9f3022013-09-23 11:33:32 -07001842void tcp_v4_init(void);
1843void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001844
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001845/* tcp_recovery.c */
Yuchung Chengdeed7be2017-01-12 22:11:32 -08001846extern void tcp_rack_mark_lost(struct sock *sk, const struct skb_mstamp *now);
Yuchung Cheng1d0833d2017-01-12 22:11:34 -08001847extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
Yuchung Chengdeed7be2017-01-12 22:11:32 -08001848 const struct skb_mstamp *xmit_time,
1849 const struct skb_mstamp *ack_time);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -08001850extern void tcp_rack_reo_timeout(struct sock *sk);
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001851
Cong Wange25f8662014-10-15 14:33:21 -07001852/*
1853 * Save and compile IPv4 options, return a pointer to it
1854 */
1855static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1856{
1857 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1858 struct ip_options_rcu *dopt = NULL;
1859
Cong Wang461b74c2014-10-15 14:33:22 -07001860 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07001861 int opt_size = sizeof(*dopt) + opt->optlen;
1862
1863 dopt = kmalloc(opt_size, GFP_ATOMIC);
1864 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1865 kfree(dopt);
1866 dopt = NULL;
1867 }
1868 }
1869 return dopt;
1870}
1871
Eric Dumazet98781962015-02-03 18:31:53 -08001872/* locally generated TCP pure ACKs have skb->truesize == 2
1873 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1874 * This is much faster than dissecting the packet to find out.
1875 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1876 */
1877static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1878{
1879 return skb->truesize == 2;
1880}
1881
1882static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1883{
1884 skb->truesize = 2;
1885}
1886
Tom Herbert473bd232016-03-07 14:11:05 -08001887static inline int tcp_inq(struct sock *sk)
1888{
1889 struct tcp_sock *tp = tcp_sk(sk);
1890 int answ;
1891
1892 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1893 answ = 0;
1894 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1895 !tp->urg_data ||
1896 before(tp->urg_seq, tp->copied_seq) ||
1897 !before(tp->urg_seq, tp->rcv_nxt)) {
1898
1899 answ = tp->rcv_nxt - tp->copied_seq;
1900
1901 /* Subtract 1, if FIN was received */
1902 if (answ && sock_flag(sk, SOCK_DONE))
1903 answ--;
1904 } else {
1905 answ = tp->urg_seq - tp->copied_seq;
1906 }
1907
1908 return answ;
1909}
1910
Tom Herbert32035582016-08-28 14:43:18 -07001911int tcp_peek_len(struct socket *sock);
1912
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07001913static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1914{
1915 u16 segs_in;
1916
1917 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1918 tp->segs_in += segs_in;
1919 if (skb->len > tcp_hdrlen(skb))
1920 tp->data_segs_in += segs_in;
1921}
1922
Eric Dumazet9caad862016-04-01 08:52:20 -07001923/*
1924 * TCP listen path runs lockless.
1925 * We forced "struct sock" to be const qualified to make sure
1926 * we don't modify one of its field by mistake.
1927 * Here, we increment sk_drops which is an atomic_t, so we can safely
1928 * make sock writable again.
1929 */
1930static inline void tcp_listendrop(const struct sock *sk)
1931{
1932 atomic_inc(&((struct sock *)sk)->sk_drops);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001933 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
Eric Dumazet9caad862016-04-01 08:52:20 -07001934}
1935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936#endif /* _TCP_H */