blob: 2b89f1ab85520d0bd426699f92f8c013c95bf35c [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
Lawrence Brakmo40304b22017-06-30 20:02:40 -070049#include <linux/bpf.h>
50#include <linux/filter.h>
51#include <linux/bpf-cgroup.h>
52
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070053extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Eric Dumazetdd24c002008-11-25 21:17:14 -080055extern struct percpu_counter tcp_orphan_count;
Joe Perches5c9f3022013-09-23 11:33:32 -070056void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070059#define MAX_TCP_OPTION_SPACE 40
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090061/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090063 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 */
65#define MAX_TCP_WINDOW 32767U
66
67/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68#define TCP_MIN_MSS 88U
69
John Heffner5d424d52006-03-20 17:53:41 -080070/* The least MTU to use for probing */
Fan Dudcd8fb82015-03-06 11:18:22 +080071#define TCP_BASE_MSS 1024
John Heffner5d424d52006-03-20 17:53:41 -080072
Fan Du05cbc0d2015-03-06 11:18:24 +080073/* probing interval, default to 10 minutes as per RFC4821 */
74#define TCP_PROBE_INTERVAL 600
75
Fan Du6b58e0a2015-03-06 11:18:23 +080076/* Specify interval when tcp mtu probing will stop */
77#define TCP_PROBE_THRESHOLD 8
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* After receiving this amount of duplicate ACKs fast retransmit starts. */
80#define TCP_FASTRETRANS_THRESH 3
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082/* Maximal number of ACKs sent quickly to accelerate slow-start. */
83#define TCP_MAX_QUICKACKS 16U
84
Gao Feng589c49c2017-04-04 21:09:48 +080085/* Maximal number of window scale according to RFC1323 */
86#define TCP_MAX_WSCALE 14U
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/* urg_data states */
89#define TCP_URG_VALID 0x0100
90#define TCP_URG_NOTYET 0x0200
91#define TCP_URG_READ 0x0400
92
93#define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
98 */
99
100#define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
105 */
106
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000107#define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
114 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000116#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
120 * initial RTO.
121 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
129 * TIME-WAIT timer.
130 */
131
132#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
133#if HZ >= 100
134#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
135#define TCP_ATO_MIN ((unsigned)(HZ/25))
136#else
137#define TCP_DELACK_MIN 4U
138#define TCP_ATO_MIN 4U
139#endif
140#define TCP_RTO_MAX ((unsigned)(120*HZ))
141#define TCP_RTO_MIN ((unsigned)(HZ/5))
Yuchung Chengbb4d9912017-07-19 15:41:26 -0700142#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000143#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000144#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
145 * used as a fallback RTO for the
146 * initial data transmission if no
147 * valid RTT sample has been acquired,
148 * most likely due to retrans in 3WHS.
149 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
152 * for local resources.
153 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
155#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
156#define TCP_KEEPALIVE_INTVL (75*HZ)
157
158#define MAX_TCP_KEEPIDLE 32767
159#define MAX_TCP_KEEPINTVL 32767
160#define MAX_TCP_KEEPCNT 127
161#define MAX_TCP_SYNCNT 127
162
163#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
166#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
167 * after this time. It should be equal
168 * (or greater than) TCP_TIMEWAIT_LEN
169 * to provide reliability equal to one
170 * provided by timewait state.
171 */
172#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
173 * timestamps. It must be less than
174 * minimal timewait lifetime.
175 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176/*
177 * TCP option
178 */
Kenjiro Nakayama105970f2014-10-20 18:15:50 +0900179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180#define TCPOPT_NOP 1 /* Padding */
181#define TCPOPT_EOL 0 /* End of options */
182#define TCPOPT_MSS 2 /* Segment size negotiating */
183#define TCPOPT_WINDOW 3 /* Window scaling */
184#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
185#define TCPOPT_SACK 5 /* SACK Block */
186#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800187#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Daniel Lee7f9b8382015-04-06 14:37:26 -0700188#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000189#define TCPOPT_EXP 254 /* Experimental */
190/* Magic number to be after the option value for sharing TCP
191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
192 */
193#define TCPOPT_FASTOPEN_MAGIC 0xF989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195/*
196 * TCP option lengths
197 */
198
199#define TCPOLEN_MSS 4
200#define TCPOLEN_WINDOW 3
201#define TCPOLEN_SACK_PERM 2
202#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800203#define TCPOLEN_MD5SIG 18
Daniel Lee7f9b8382015-04-06 14:37:26 -0700204#define TCPOLEN_FASTOPEN_BASE 2
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000205#define TCPOLEN_EXP_FASTOPEN_BASE 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
207/* But this is what stacks really send out. */
208#define TCPOLEN_TSTAMP_ALIGNED 12
209#define TCPOLEN_WSCALE_ALIGNED 4
210#define TCPOLEN_SACKPERM_ALIGNED 4
211#define TCPOLEN_SACK_BASE 2
212#define TCPOLEN_SACK_BASE_ALIGNED 4
213#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800214#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700215#define TCPOLEN_MSS_ALIGNED 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217/* Flags in tp->nonagle */
218#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800220#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000222/* TCP thin-stream limits */
223#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
224
Jörg Thalheim21603fc2016-01-27 22:54:06 +0100225/* TCP initial congestion window as per rfc6928 */
David S. Miller442b9632011-02-02 17:05:11 -0800226#define TCP_INIT_CWND 10
227
Yuchung Chengcf60af02012-07-19 06:43:09 +0000228/* Bit Flags for sysctl_tcp_fastopen */
229#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000230#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000231#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af02012-07-19 06:43:09 +0000232
Jerry Chu10467162012-08-31 12:29:11 +0000233/* Accept SYN data w/o any cookie option */
234#define TFO_SERVER_COOKIE_NOT_REQD 0x200
235
236/* Force enable TFO on all listeners, i.e., not requiring the
Yuchung Chengcebc5cb2016-08-22 17:17:54 -0700237 * TCP_FASTOPEN socket option.
Jerry Chu10467162012-08-31 12:29:11 +0000238 */
239#define TFO_SERVER_WO_SOCKOPT1 0x400
Jerry Chu10467162012-08-31 12:29:11 +0000240
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242/* sysctl variables for tcp */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000243extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244extern int sysctl_tcp_retrans_collapse;
245extern int sysctl_tcp_stdurg;
246extern int sysctl_tcp_rfc1337;
247extern int sysctl_tcp_abort_on_overflow;
248extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249extern int sysctl_tcp_fack;
250extern int sysctl_tcp_reordering;
Eric Dumazetdca145f2014-10-27 21:45:24 -0700251extern int sysctl_tcp_max_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700253extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254extern int sysctl_tcp_wmem[3];
255extern int sysctl_tcp_rmem[3];
256extern int sysctl_tcp_app_win;
257extern int sysctl_tcp_adv_win_scale;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258extern int sysctl_tcp_frto;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260extern int sysctl_tcp_moderate_rcvbuf;
261extern int sysctl_tcp_tso_win_divisor;
Rick Jones15d99e02006-03-20 22:40:29 -0800262extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700263extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000264extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000265extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000266extern int sysctl_tcp_early_retrans;
Yuchung Chenga0370b32017-01-12 22:11:36 -0800267extern int sysctl_tcp_recovery;
268#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
269
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000270extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200271extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700272extern int sysctl_tcp_min_tso_segs;
Yuchung Chengf6722582015-10-16 21:57:42 -0700273extern int sysctl_tcp_min_rtt_wlen;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800274extern int sysctl_tcp_autocorking;
Neal Cardwell032ee422015-02-06 16:04:38 -0500275extern int sysctl_tcp_invalid_ratelimit;
Eric Dumazet43e122b2015-08-21 17:38:02 -0700276extern int sysctl_tcp_pacing_ss_ratio;
277extern int sysctl_tcp_pacing_ca_ratio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Eric Dumazet8d987e52010-11-09 23:24:26 +0000279extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800280extern struct percpu_counter tcp_sockets_allocated;
Eric Dumazet06044752017-06-07 13:29:12 -0700281extern unsigned long tcp_memory_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700283/* optimized version of sk_under_memory_pressure() for TCP sockets */
284static inline bool tcp_under_memory_pressure(const struct sock *sk)
285{
Johannes Weinerbaac50b2016-01-14 15:21:17 -0800286 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
287 mem_cgroup_under_socket_pressure(sk->sk_memcg))
Johannes Weinere8056052016-01-14 15:21:14 -0800288 return true;
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700289
290 return tcp_memory_pressure;
291}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 * The next routines deal with comparing 32 bit unsigned ints
294 * and worry about wraparound (automatic with unsigned arithmetic).
295 */
296
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000297static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800299 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800301#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000304static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
306 return seq3 - seq2 >= seq1 - seq2;
307}
308
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800309static inline bool tcp_out_of_memory(struct sock *sk)
310{
311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
313 return true;
314 return false;
315}
316
Eric Dumazeta6c5ea42015-05-15 12:39:26 -0700317void sk_forced_mem_schedule(struct sock *sk, int size);
318
David S. Millerad1af0f2010-08-25 02:27:49 -0700319static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700320{
David S. Millerad1af0f2010-08-25 02:27:49 -0700321 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
322 int orphans = percpu_counter_read_positive(ocp);
323
324 if (orphans << shift > sysctl_tcp_max_orphans) {
325 orphans = percpu_counter_sum_positive(ocp);
326 if (orphans << shift > sysctl_tcp_max_orphans)
327 return true;
328 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700329 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700330}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
Joe Perches5c9f3022013-09-23 11:33:32 -0700332bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800333
Florian Westphala0f82f62009-04-19 09:43:48 +0000334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335extern struct proto tcp_prot;
336
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700337#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700338#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700339#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000340#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Joe Perches5c9f3022013-09-23 11:33:32 -0700342void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000343
Joe Perches5c9f3022013-09-23 11:33:32 -0700344void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Joe Perches5c9f3022013-09-23 11:33:32 -0700346void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Joe Perches5c9f3022013-09-23 11:33:32 -0700348void tcp_v4_early_demux(struct sk_buff *skb);
349int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
Joe Perches5c9f3022013-09-23 11:33:32 -0700351int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800352int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Tom Herbert306b13e2017-07-28 16:22:41 -0700353int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700354int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
355 int flags);
Tom Herbert306b13e2017-07-28 16:22:41 -0700356int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
357 size_t size, int flags);
Dave Watsone3b56162017-06-14 11:37:26 -0700358ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
359 size_t size, int flags);
Joe Perches5c9f3022013-09-23 11:33:32 -0700360void tcp_release_cb(struct sock *sk);
361void tcp_wfree(struct sk_buff *skb);
362void tcp_write_timer_handler(struct sock *sk);
363void tcp_delack_timer_handler(struct sock *sk);
364int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700365int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700366void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
Matvejchikov Ilyae42e24c2017-07-24 16:02:12 +0400367 const struct tcphdr *th);
Joe Perches5c9f3022013-09-23 11:33:32 -0700368void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700369int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
370void tcp_twsk_destructor(struct sock *sk);
371ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
372 struct pipe_inode_info *pipe, size_t len,
373 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800374
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700375static inline void tcp_dec_quickack_mode(struct sock *sk,
376 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700378 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700379
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700380 if (icsk->icsk_ack.quick) {
381 if (pkts >= icsk->icsk_ack.quick) {
382 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700383 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700384 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700385 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700386 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
388}
389
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700390#define TCP_ECN_OK 1
391#define TCP_ECN_QUEUE_CWR 2
392#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000393#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700394
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000395enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 TCP_TW_SUCCESS = 0,
397 TCP_TW_RST = 1,
398 TCP_TW_ACK = 2,
399 TCP_TW_SYN = 3
400};
401
402
Joe Perches5c9f3022013-09-23 11:33:32 -0700403enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
404 struct sk_buff *skb,
405 const struct tcphdr *th);
406struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazet52452c52015-03-19 19:04:19 -0700407 struct request_sock *req, bool fastopen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700408int tcp_child_process(struct sock *parent, struct sock *child,
409 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400410void tcp_enter_loss(struct sock *sk);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800411void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
Joe Perches5c9f3022013-09-23 11:33:32 -0700412void tcp_clear_retrans(struct tcp_sock *tp);
413void tcp_update_metrics(struct sock *sk);
414void tcp_init_metrics(struct sock *sk);
415void tcp_metrics_init(void);
Soheil Hassas Yeganehd82bae12017-03-15 16:30:45 -0400416bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700417void tcp_disable_fack(struct tcp_sock *tp);
418void tcp_close(struct sock *sk, long timeout);
419void tcp_init_sock(struct sock *sk);
420unsigned int tcp_poll(struct file *file, struct socket *sock,
421 struct poll_table_struct *wait);
422int tcp_getsockopt(struct sock *sk, int level, int optname,
423 char __user *optval, int __user *optlen);
424int tcp_setsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, unsigned int optlen);
426int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000427 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700428int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000429 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700430void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700431void tcp_syn_ack_timeout(const struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800432int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
433 int flags, int *addr_len);
Eric Dumazeteed29f12017-06-07 10:34:36 -0700434void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700435 struct tcp_options_received *opt_rx,
436 int estab, struct tcp_fastopen_cookie *foc);
437const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439/*
440 * TCP v4 functions exported for the inet6 API
441 */
442
Joe Perches5c9f3022013-09-23 11:33:32 -0700443void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400444void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet9cf74902016-02-02 19:31:12 -0800445void tcp_req_err(struct sock *sk, u32 seq, bool abort);
Joe Perches5c9f3022013-09-23 11:33:32 -0700446int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700447struct sock *tcp_create_openreq_child(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -0700448 struct request_sock *req,
449 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100450void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Eric Dumazet0c271712015-09-29 07:42:48 -0700451struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700452 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700453 struct dst_entry *dst,
454 struct request_sock *req_unhash,
455 bool *own_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700456int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
457int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
458int tcp_connect(struct sock *sk);
Eric Dumazetb3d05142016-04-13 22:05:39 -0700459enum tcp_synack_type {
460 TCP_SYNACK_NORMAL,
461 TCP_SYNACK_FASTOPEN,
462 TCP_SYNACK_COOKIE,
463};
Eric Dumazet5d062de2015-09-25 07:39:19 -0700464struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
Joe Perches5c9f3022013-09-23 11:33:32 -0700465 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700466 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -0700467 enum tcp_synack_type synack_type);
Joe Perches5c9f3022013-09-23 11:33:32 -0700468int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000470void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000471int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000472void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/* From syncookies.c */
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700475struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
476 struct request_sock *req,
Eric Dumazet84b114b2017-05-05 06:56:54 -0700477 struct dst_entry *dst, u32 tsoff);
Joe Perches5c9f3022013-09-23 11:33:32 -0700478int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
479 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700480struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400481#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200482
Eric Dumazet63262312014-03-19 21:02:21 -0700483/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200484 * This counter is used both as a hash input and partially encoded into
485 * the cookie value. A cookie is only validated further if the delta
486 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200488 * the counter advances immediately after a cookie is generated).
489 */
Eric Dumazet264ea102015-05-14 14:26:56 -0700490#define MAX_SYNCOOKIE_AGE 2
491#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
492#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
493
494/* syncookies: remember time of last synqueue overflow
495 * But do not dirty this field too often (once per second is enough)
Eric Dumazet3f684b42015-09-29 07:42:49 -0700496 * It is racy as we do not hold a lock, but race is very minor.
Eric Dumazet264ea102015-05-14 14:26:56 -0700497 */
Eric Dumazet3f684b42015-09-29 07:42:49 -0700498static inline void tcp_synq_overflow(const struct sock *sk)
Eric Dumazet264ea102015-05-14 14:26:56 -0700499{
500 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
501 unsigned long now = jiffies;
502
503 if (time_after(now, last_overflow + HZ))
504 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
505}
506
507/* syncookies: no recent synqueue overflow on this listening socket? */
508static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
509{
510 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
511
512 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
513}
Florian Westphal8c27bd72013-09-20 22:32:55 +0200514
515static inline u32 tcp_cookie_time(void)
516{
Eric Dumazet63262312014-03-19 21:02:21 -0700517 u64 val = get_jiffies_64();
518
Eric Dumazet264ea102015-05-14 14:26:56 -0700519 do_div(val, TCP_SYNCOOKIE_PERIOD);
Eric Dumazet63262312014-03-19 21:02:21 -0700520 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200521}
522
Joe Perches5c9f3022013-09-23 11:33:32 -0700523u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
524 u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700525__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazet9a568de2017-05-16 14:00:14 -0700526u64 cookie_init_timestamp(struct request_sock *req);
Eric Dumazetf9301032017-06-07 10:34:37 -0700527bool cookie_timestamp_decode(const struct net *net,
528 struct tcp_options_received *opt);
Florian Westphalf1673382014-11-03 17:35:02 +0100529bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100530 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700531
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800532/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700533int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
534 u32 cookie);
535struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100536
Joe Perches5c9f3022013-09-23 11:33:32 -0700537u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
538 const struct tcphdr *th, u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700539__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400540#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541/* tcp_output.c */
542
Neal Cardwell1b3878c2016-09-19 23:39:18 -0400543u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
544 int min_tso_segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700545void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
546 int nonagle);
547bool tcp_may_send_now(struct sock *sk);
Eric Dumazet10d3be52016-04-21 10:55:23 -0700548int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
549int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700550void tcp_retransmit_timer(struct sock *sk);
551void tcp_xmit_retransmit_queue(struct sock *);
552void tcp_simple_retransmit(struct sock *);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800553void tcp_enter_recovery(struct sock *sk, bool ece_ack);
Joe Perches5c9f3022013-09-23 11:33:32 -0700554int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300555int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
Joe Perches5c9f3022013-09-23 11:33:32 -0700557void tcp_send_probe0(struct sock *);
558void tcp_send_partial(struct sock *);
Eric Dumazete520af42015-05-06 14:26:25 -0700559int tcp_write_wakeup(struct sock *, int mib);
Joe Perches5c9f3022013-09-23 11:33:32 -0700560void tcp_send_fin(struct sock *sk);
561void tcp_send_active_reset(struct sock *sk, gfp_t priority);
562int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700563void tcp_push_one(struct sock *, unsigned int mss_now);
564void tcp_send_ack(struct sock *sk);
565void tcp_send_delayed_ack(struct sock *sk);
566void tcp_send_loss_probe(struct sock *sk);
567bool tcp_schedule_loss_probe(struct sock *sk);
Martin KaFai Laucfea5a62016-04-19 22:39:29 -0700568void tcp_skb_collapse_tstamp(struct sk_buff *skb,
569 const struct sk_buff *next_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
David S. Millera762a982005-07-05 15:18:51 -0700571/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700572void tcp_rearm_rto(struct sock *sk);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700573void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700574void tcp_reset(struct sock *sk);
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -0700575void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
Eric Dumazete3e17b72016-02-06 11:16:28 -0800576void tcp_fin(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700579void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700580static inline void tcp_clear_xmit_timers(struct sock *sk)
581{
Eric Dumazet218af592017-05-16 04:24:36 -0700582 hrtimer_cancel(&tcp_sk(sk)->pacing_timer);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700583 inet_csk_clear_xmit_timers(sk);
584}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Joe Perches5c9f3022013-09-23 11:33:32 -0700586unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
587unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000588
589/* Bound MSS / TSO packet size with the half of the window */
590static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
591{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700592 int cutoff;
593
594 /* When peer uses tiny windows, there is no use in packetizing
595 * to sub-MSS pieces for the sake of SWS or making sure there
596 * are enough packets in the pipe for fast recovery.
597 *
598 * On the other hand, for extremely large MSS devices, handling
599 * smaller than MSS windows in this way does make sense.
600 */
Seymour, Shane M2631b792016-06-28 23:06:48 +0000601 if (tp->max_window > TCP_MSS_DEFAULT)
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700602 cutoff = (tp->max_window >> 1);
603 else
604 cutoff = tp->max_window;
605
606 if (cutoff && pktsize > cutoff)
607 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000608 else
609 return pktsize;
610}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300612/* tcp.c */
Eric Dumazet0df48c22015-04-28 15:28:17 -0700613void tcp_get_info(struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615/* Read 'sendfile()'-style from a TCP socket */
Joe Perches5c9f3022013-09-23 11:33:32 -0700616int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
617 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
Joe Perches5c9f3022013-09-23 11:33:32 -0700619void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Joe Perches5c9f3022013-09-23 11:33:32 -0700621int tcp_mtu_to_mss(struct sock *sk, int pmtu);
622int tcp_mss_to_mtu(struct sock *sk, int mss);
623void tcp_mtup_init(struct sock *sk);
624void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800625
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000626static inline void tcp_bound_rto(const struct sock *sk)
627{
628 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
629 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
630}
631
632static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
633{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800634 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000635}
636
Satoru SATOH0c266892009-05-04 11:11:01 -0700637/* Compute the actual rto_min value */
638static inline u32 tcp_rto_min(struct sock *sk)
639{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400640 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700641 u32 rto_min = TCP_RTO_MIN;
642
643 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
644 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
645 return rto_min;
646}
647
Eric Dumazet740b0f12014-02-26 14:02:48 -0800648static inline u32 tcp_rto_min_us(struct sock *sk)
649{
650 return jiffies_to_usecs(tcp_rto_min(sk));
651}
652
Daniel Borkmann81164412015-01-05 23:57:48 +0100653static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
654{
655 return dst_metric_locked(dst, RTAX_CC_ALGO);
656}
657
Yuchung Chengf6722582015-10-16 21:57:42 -0700658/* Minimum RTT in usec. ~0 means not available. */
659static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
660{
Neal Cardwell64033892016-09-19 23:39:10 -0400661 return minmax_get(&tp->rtt_min);
Yuchung Chengf6722582015-10-16 21:57:42 -0700662}
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664/* Compute the actual receive window we are currently advertising.
665 * Rcv_nxt can be after the window if our peer push more data
666 * than the offered window.
667 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800668static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
671
672 if (win < 0)
673 win = 0;
674 return (u32) win;
675}
676
677/* Choose a new window, without checks for shrinking, and without
678 * scaling applied to the result. The caller does these things
679 * if necessary. This is a "raw" window selection.
680 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700681u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Pavel Emelyanovee995282012-04-19 03:40:39 +0000683void tcp_send_window_probe(struct sock *sk);
684
Eric Dumazetec66eda2017-05-16 14:00:01 -0700685/* TCP uses 32bit jiffies to save some space.
686 * Note that this is different from tcp_time_stamp, which
687 * historically has been the same until linux-4.13.
688 */
689#define tcp_jiffies32 ((u32)jiffies)
690
Eric Dumazet9a568de2017-05-16 14:00:14 -0700691/*
692 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
693 * It is no longer tied to jiffies, but to 1 ms clock.
694 * Note: double check if you want to use tcp_jiffies32 instead of this.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700696#define TCP_TS_HZ 1000
697
698static inline u64 tcp_clock_ns(void)
699{
700 return local_clock();
701}
702
703static inline u64 tcp_clock_us(void)
704{
705 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
706}
707
708/* This should only be used in contexts where tp->tcp_mstamp is up to date */
709static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
710{
711 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
712}
713
714/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
715static inline u32 tcp_time_stamp_raw(void)
716{
717 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
718}
719
720
721/* Refresh 1us clock of a TCP socket,
722 * ensuring monotically increasing values.
723 */
724static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
725{
726 u64 val = tcp_clock_us();
727
728 if (val > tp->tcp_mstamp)
729 tp->tcp_mstamp = val;
730}
731
732static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
733{
734 return max_t(s64, t1 - t0, 0);
735}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700737static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
738{
Eric Dumazet9a568de2017-05-16 14:00:14 -0700739 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700740}
741
742
Changli Gaoa3433f32010-06-12 14:01:43 +0000743#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
744
745#define TCPHDR_FIN 0x01
746#define TCPHDR_SYN 0x02
747#define TCPHDR_RST 0x04
748#define TCPHDR_PSH 0x08
749#define TCPHDR_ACK 0x10
750#define TCPHDR_URG 0x20
751#define TCPHDR_ECE 0x40
752#define TCPHDR_CWR 0x80
753
Daniel Borkmann49213552015-05-19 21:04:22 +0200754#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
755
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800756/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700757 * TCP per-packet control information to the transmission code.
758 * We also store the host-order sequence numbers in here too.
759 * This is 44 bytes if IPV6 is enabled.
760 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 */
762struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 __u32 seq; /* Starting sequence number */
764 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700765 union {
766 /* Note : tcp_tw_isn is used in input path only
767 * (isn chosen by tcp_timewait_state_process())
768 *
Eric Dumazetf69ad292015-06-11 09:15:18 -0700769 * tcp_gso_segs/size are used in write queue only,
770 * cf tcp_skb_pcount()/tcp_skb_mss()
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700771 */
772 __u32 tcp_tw_isn;
Eric Dumazetf69ad292015-06-11 09:15:18 -0700773 struct {
774 u16 tcp_gso_segs;
775 u16 tcp_gso_size;
776 };
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700777 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400778 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 __u8 sacked; /* State flags for SACK/FACK. */
781#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
782#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
783#define TCPCB_LOST 0x04 /* SKB is lost */
784#define TCPCB_TAGBITS 0x07 /* All tag bits */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400785#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400787#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
788 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000790 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
Soheil Hassas Yeganeh6b084922016-04-02 23:08:08 -0400791 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700792 eor:1, /* Is skb MSG_EOR marked? */
793 unused:6;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700795 union {
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700796 struct {
Yuchung Chengb9f64822016-09-19 23:39:14 -0400797 /* There is space for up to 24 bytes */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400798 __u32 in_flight:30,/* Bytes in flight at transmit */
799 is_app_limited:1, /* cwnd not fully used? */
800 unused:1;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400801 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
802 __u32 delivered;
803 /* start of send pipeline phase */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700804 u64 first_tx_mstamp;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400805 /* when we reached the "delivered" count */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700806 u64 delivered_mstamp;
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700807 } tx; /* only used for outgoing skbs */
808 union {
809 struct inet_skb_parm h4;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700810#if IS_ENABLED(CONFIG_IPV6)
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700811 struct inet6_skb_parm h6;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700812#endif
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700813 } header; /* For incoming skbs */
814 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815};
816
817#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
818
Eric Dumazet870c3152014-10-17 09:17:20 -0700819
Eric Dumazet815afe12014-10-18 08:34:37 -0700820#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700821/* This is the variant of inet6_iif() that must be used by TCP,
822 * as TCP moves IP6CB into a different location in skb->cb[]
823 */
824static inline int tcp_v6_iif(const struct sk_buff *skb)
825{
David Aherna04a4802016-10-16 20:02:52 -0700826 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
David Ahern74b20582016-05-10 11:19:50 -0700827
828 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
Eric Dumazet870c3152014-10-17 09:17:20 -0700829}
Eric Dumazet815afe12014-10-18 08:34:37 -0700830#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700831
David Aherna04a4802016-10-16 20:02:52 -0700832/* TCP_SKB_CB reference means this can not be used from early demux */
833static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
834{
835#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
836 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
David Ahernda967862016-11-02 12:08:25 -0700837 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
David Aherna04a4802016-10-16 20:02:52 -0700838 return true;
839#endif
840 return false;
841}
842
David Ahern3fa6f612017-08-07 08:44:17 -0700843/* TCP_SKB_CB reference means this can not be used from early demux */
844static inline int tcp_v4_sdif(struct sk_buff *skb)
845{
846#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
847 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
848 return TCP_SKB_CB(skb)->header.h4.iif;
849#endif
850 return 0;
851}
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853/* Due to TSO, an SKB can be composed of multiple actual
854 * packets. To keep these tracked properly, we use this.
855 */
856static inline int tcp_skb_pcount(const struct sk_buff *skb)
857{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700858 return TCP_SKB_CB(skb)->tcp_gso_segs;
859}
860
861static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
862{
863 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
864}
865
866static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
867{
868 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869}
870
Eric Dumazetf69ad292015-06-11 09:15:18 -0700871/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872static inline int tcp_skb_mss(const struct sk_buff *skb)
873{
Eric Dumazetf69ad292015-06-11 09:15:18 -0700874 return TCP_SKB_CB(skb)->tcp_gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700877static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
878{
879 return likely(!TCP_SKB_CB(skb)->eor);
880}
881
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700882/* Events passed to congestion control interface */
883enum tcp_ca_event {
884 CA_EVENT_TX_START, /* first transmit when no packets in flight */
885 CA_EVENT_CWND_RESTART, /* congestion window restart */
886 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700887 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200888 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
889 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
890 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
891 CA_EVENT_NON_DELAYED_ACK,
Florian Westphal7354c8c2014-09-26 22:37:34 +0200892};
893
Florian Westphal98900922014-09-26 22:37:35 +0200894/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200895enum tcp_ca_ack_event_flags {
Florian Westphal573aeb02017-07-30 03:57:22 +0200896 CA_ACK_WIN_UPDATE = (1 << 0), /* ACK updated window */
897 CA_ACK_ECE = (1 << 1), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700898};
899
900/*
901 * Interface for adding new TCP congestion control handlers
902 */
903#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800904#define TCP_CA_MAX 128
905#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
906
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100907#define TCP_CA_UNSPEC 0
908
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200909/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700910#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200911/* Requires ECN/ECT set on all packets */
912#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700913
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700914union tcp_cc_info;
915
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700916struct ack_sample {
917 u32 pkts_acked;
918 s32 rtt_us;
Lawrence Brakmo6f094b92016-06-08 21:16:44 -0700919 u32 in_flight;
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700920};
921
Yuchung Chengb9f64822016-09-19 23:39:14 -0400922/* A rate sample measures the number of (original/retransmitted) data
923 * packets delivered "delivered" over an interval of time "interval_us".
924 * The tcp_rate.c code fills in the rate sample, and congestion
925 * control modules that define a cong_control function to run at the end
926 * of ACK processing can optionally chose to consult this sample when
927 * setting cwnd and pacing rate.
928 * A sample is invalid if "delivered" or "interval_us" is negative.
929 */
930struct rate_sample {
Eric Dumazet9a568de2017-05-16 14:00:14 -0700931 u64 prior_mstamp; /* starting timestamp for interval */
Yuchung Chengb9f64822016-09-19 23:39:14 -0400932 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
933 s32 delivered; /* number of packets delivered over interval */
934 long interval_us; /* time for tp->delivered to incr "delivered" */
935 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
936 int losses; /* number of packets marked lost upon ACK */
937 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
938 u32 prior_in_flight; /* in flight before this ACK */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400939 bool is_app_limited; /* is sample from packet with bubble in pipe? */
Yuchung Chengb9f64822016-09-19 23:39:14 -0400940 bool is_retrans; /* is sample from retransmission? */
941};
942
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700943struct tcp_congestion_ops {
944 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100945 u32 key;
946 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700947
948 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300949 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700950 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300951 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700952
953 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300954 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700955 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700956 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700957 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300958 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700959 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300960 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +0200961 /* call when ack arrives (optional) */
962 void (*in_ack_event)(struct sock *sk, u32 flags);
Anmol Sarma1e0ce2a2017-06-03 17:40:54 +0530963 /* new value of cwnd after loss (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300964 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700965 /* hook for packet ack accounting (optional) */
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700966 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
Neal Cardwelled6e7262016-09-19 23:39:17 -0400967 /* suggest number of segments for each skb to transmit (optional) */
968 u32 (*tso_segs_goal)(struct sock *sk);
Yuchung Cheng77bfc172016-09-19 23:39:20 -0400969 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
970 u32 (*sndbuf_expand)(struct sock *sk);
Yuchung Chengc0402762016-09-19 23:39:21 -0400971 /* call when packets are delivered to update cwnd and pacing rate,
972 * after all the ca_state processing. (optional)
973 */
974 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300975 /* get info for inet_diag (optional) */
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700976 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
977 union tcp_cc_info *info);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700978
979 char name[TCP_CA_NAME_MAX];
980 struct module *owner;
981};
982
Joe Perches5c9f3022013-09-23 11:33:32 -0700983int tcp_register_congestion_control(struct tcp_congestion_ops *type);
984void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700985
Florian Westphal55d86942014-09-26 22:37:32 +0200986void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700987void tcp_init_congestion_control(struct sock *sk);
988void tcp_cleanup_congestion_control(struct sock *sk);
989int tcp_set_default_congestion_control(const char *name);
990void tcp_get_default_congestion_control(char *name);
991void tcp_get_available_congestion_control(char *buf, size_t len);
992void tcp_get_allowed_congestion_control(char *buf, size_t len);
993int tcp_set_allowed_congestion_control(char *allowed);
Lawrence Brakmo91b5b212017-06-30 20:02:49 -0700994int tcp_set_congestion_control(struct sock *sk, const char *name, bool load);
995void tcp_reinit_congestion_control(struct sock *sk,
996 const struct tcp_congestion_ops *ca);
Neal Cardwelle73ebb082015-01-28 20:01:35 -0500997u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
998void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700999
Joe Perches5c9f3022013-09-23 11:33:32 -07001000u32 tcp_reno_ssthresh(struct sock *sk);
Florian Westphale9799182016-11-21 14:18:38 +01001001u32 tcp_reno_undo_cwnd(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -07001002void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -07001003extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001004
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001005struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
Daniel Borkmannc3a8d942015-08-31 15:58:47 +02001006u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
Daniel Borkmannea697632015-01-05 23:57:47 +01001007#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001008char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +01001009#else
1010static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1011{
1012 return NULL;
1013}
1014#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001015
Daniel Borkmann30e502a2014-09-26 22:37:33 +02001016static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1017{
1018 const struct inet_connection_sock *icsk = inet_csk(sk);
1019
1020 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1021}
1022
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001023static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001024{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001025 struct inet_connection_sock *icsk = inet_csk(sk);
1026
1027 if (icsk->icsk_ca_ops->set_state)
1028 icsk->icsk_ca_ops->set_state(sk, ca_state);
1029 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001030}
1031
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001032static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001033{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001034 const struct inet_connection_sock *icsk = inet_csk(sk);
1035
1036 if (icsk->icsk_ca_ops->cwnd_event)
1037 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001038}
1039
Yuchung Chengb9f64822016-09-19 23:39:14 -04001040/* From tcp_rate.c */
1041void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1042void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1043 struct rate_sample *rs);
1044void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
Eric Dumazet88d5c652017-04-25 10:15:37 -07001045 struct rate_sample *rs);
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001046void tcp_rate_check_app_limited(struct sock *sk);
Yuchung Chengb9f64822016-09-19 23:39:14 -04001047
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001048/* These functions determine how the current flow behaves in respect of SACK
1049 * handling. SACK is negotiated with the peer, and therefore it can vary
1050 * between different flows.
1051 *
1052 * tcp_is_sack - SACK enabled
1053 * tcp_is_reno - No SACK
1054 * tcp_is_fack - FACK enabled, implies SACK enabled
1055 */
1056static inline int tcp_is_sack(const struct tcp_sock *tp)
1057{
1058 return tp->rx_opt.sack_ok;
1059}
1060
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001061static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001062{
1063 return !tcp_is_sack(tp);
1064}
1065
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001066static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001067{
Vijay Subramanianab562222011-12-20 13:23:24 +00001068 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001069}
1070
1071static inline void tcp_enable_fack(struct tcp_sock *tp)
1072{
Vijay Subramanianab562222011-12-20 13:23:24 +00001073 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001074}
1075
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001076static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1077{
1078 return tp->sacked_out + tp->lost_out;
1079}
1080
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081/* This determines how many packets are "in the network" to the best
1082 * of our knowledge. In many cases it is conservative, but where
1083 * detailed information is available from the receiver (via SACK
1084 * blocks etc.) we can make more aggressive calculations.
1085 *
1086 * Use this for decisions involving congestion control, use just
1087 * tp->packets_out to determine if the send queue is empty or not.
1088 *
1089 * Read this equation as:
1090 *
1091 * "Packets sent once on transmission queue" MINUS
1092 * "Packets left network, but not honestly ACKed yet" PLUS
1093 * "Packets fast retransmitted"
1094 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001095static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001097 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001100#define TCP_INFINITE_SSTHRESH 0x7fffffff
1101
Yuchung Cheng071d5082015-07-09 13:16:29 -07001102static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1103{
Yuchung Cheng76174002015-07-09 13:16:30 -07001104 return tp->snd_cwnd < tp->snd_ssthresh;
Yuchung Cheng071d5082015-07-09 13:16:29 -07001105}
1106
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001107static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1108{
1109 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1110}
1111
Yuchung Cheng684bad12012-09-02 17:38:04 +00001112static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1113{
1114 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1115 (1 << inet_csk(sk)->icsk_ca_state);
1116}
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +00001119 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 * ssthresh.
1121 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001122static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001124 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001125
Yuchung Cheng684bad12012-09-02 17:38:04 +00001126 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 return tp->snd_ssthresh;
1128 else
1129 return max(tp->snd_ssthresh,
1130 ((tp->snd_cwnd >> 1) +
1131 (tp->snd_cwnd >> 2)));
1132}
1133
Ilpo Järvinenb9c45952007-07-27 16:36:17 +03001134/* Use define here intentionally to get WARN_ON location shown at the caller */
1135#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
Christoph Paasch5ee2c942014-07-14 16:58:32 +02001137void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001138__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Neal Cardwell6b5a5c02011-11-21 17:15:14 +00001140/* The maximum number of MSS of available cwnd for which TSO defers
1141 * sending if not using sysctl_tcp_tso_win_divisor.
1142 */
1143static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1144{
1145 return 3;
1146}
1147
Ilpo Järvinen90840de2007-12-31 04:48:41 -08001148/* Returns end sequence number of the receiver's advertised window */
1149static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1150{
1151 return tp->snd_una + tp->snd_wnd;
1152}
Eric Dumazete114a712014-04-30 11:58:13 -07001153
1154/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1155 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001156 * it was fully used previously. And that's exactly what we do in
1157 * congestion avoidance mode. But in slow start we allow cwnd to grow
1158 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001159 * Example :
1160 * cwnd is 10 (IW10), but application sends 9 frames.
1161 * We allow cwnd to reach 18 when all frames are ACKed.
1162 * This check is safe because it's as aggressive as slow start which already
1163 * risks 100% overshoot. The advantage is that we discourage application to
1164 * either send more filler packets or data to artificially blow up the cwnd
1165 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001166 */
Eric Dumazet24901552014-05-02 21:18:05 -07001167static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001168{
1169 const struct tcp_sock *tp = tcp_sk(sk);
1170
Neal Cardwellca8a2262014-05-22 10:41:08 -04001171 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
Yuchung Cheng071d5082015-07-09 13:16:29 -07001172 if (tcp_in_slow_start(tp))
Neal Cardwellca8a2262014-05-22 10:41:08 -04001173 return tp->snd_cwnd < 2 * tp->max_packets_out;
1174
1175 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001176}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001177
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001178/* Something is really bad, we could not queue an additional packet,
1179 * because qdisc is full or receiver sent a 0 window.
1180 * We do not want to add fuel to the fire, or abort too early,
1181 * so make sure the timer we arm now is at least 200ms in the future,
1182 * regardless of current icsk_rto value (as it could be ~2ms)
1183 */
1184static inline unsigned long tcp_probe0_base(const struct sock *sk)
1185{
1186 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1187}
1188
1189/* Variant of inet_csk_rto_backoff() used for zero window probes */
1190static inline unsigned long tcp_probe0_when(const struct sock *sk,
1191 unsigned long max_when)
1192{
1193 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1194
1195 return (unsigned long)min_t(u64, when, max_when);
1196}
1197
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001198static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001200 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001201 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001202 tcp_probe0_base(sk), TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001205static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
1207 tp->snd_wl1 = seq;
1208}
1209
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001210static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 tp->snd_wl1 = seq;
1213}
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215/*
1216 * Calculate(/check) TCP checksum
1217 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001218static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1219 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
1221 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1222}
1223
Al Virob51655b2006-11-14 21:40:42 -08001224static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Herbert Xufb286bb2005-11-10 13:01:24 -08001226 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
1228
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001229static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230{
Herbert Xu60476372007-04-09 11:59:39 -07001231 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 __tcp_checksum_complete(skb);
1233}
1234
Eric Dumazetc9c33212016-08-27 07:37:54 -07001235bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
Eric Dumazetac6e7802016-11-10 13:12:35 -08001236int tcp_filter(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
1238#undef STATE_TRACE
1239
1240#ifdef STATE_TRACE
1241static const char *statename[]={
1242 "Unused","Established","Syn Sent","Syn Recv",
1243 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1244 "Close Wait","Last ACK","Listen","Closing"
1245};
1246#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001247void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Joe Perches5c9f3022013-09-23 11:33:32 -07001249void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09001251int tcp_abort(struct sock *sk, int err);
1252
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001253static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
1255 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 rx_opt->num_sacks = 0;
1257}
1258
Joe Perches5c9f3022013-09-23 11:33:32 -07001259u32 tcp_default_init_rwnd(u32 mss);
Eric Dumazet6f021c62015-08-21 12:30:00 -07001260void tcp_cwnd_restart(struct sock *sk, s32 delta);
1261
1262static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1263{
Wei Wang1b1fc3f2017-05-05 12:53:23 -07001264 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
Eric Dumazet6f021c62015-08-21 12:30:00 -07001265 struct tcp_sock *tp = tcp_sk(sk);
1266 s32 delta;
1267
Wei Wang1b1fc3f2017-05-05 12:53:23 -07001268 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1269 ca_ops->cong_control)
Eric Dumazet6f021c62015-08-21 12:30:00 -07001270 return;
Eric Dumazetd635fbe2017-05-16 14:00:03 -07001271 delta = tcp_jiffies32 - tp->lsndtime;
Eric Dumazet6f021c62015-08-21 12:30:00 -07001272 if (delta > inet_csk(sk)->icsk_rto)
1273 tcp_cwnd_restart(sk, delta);
1274}
Yuchung Cheng85f16522013-06-11 15:35:32 -07001275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001277void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1278 __u32 *window_clamp, int wscale_ok,
1279 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
1281static inline int tcp_win_from_space(int space)
1282{
Gao Fengc4836742017-03-24 07:05:12 +08001283 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1284
1285 return tcp_adv_win_scale <= 0 ?
1286 (space>>(-tcp_adv_win_scale)) :
1287 space - (space>>tcp_adv_win_scale);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288}
1289
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001290/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291static inline int tcp_space(const struct sock *sk)
1292{
1293 return tcp_win_from_space(sk->sk_rcvbuf -
1294 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001295}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297static inline int tcp_full_space(const struct sock *sk)
1298{
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001299 return tcp_win_from_space(sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300}
1301
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001302extern void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -07001303 const struct sock *sk_listener,
1304 const struct dst_entry *dst);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001305
Joe Perches5c9f3022013-09-23 11:33:32 -07001306void tcp_enter_memory_pressure(struct sock *sk);
Eric Dumazet06044752017-06-07 13:29:12 -07001307void tcp_leave_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1310{
Nikolay Borisovb840d152016-01-07 16:38:45 +02001311 struct net *net = sock_net((struct sock *)tp);
1312
1313 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
1315
1316static inline int keepalive_time_when(const struct tcp_sock *tp)
1317{
Nikolay Borisov13b287e2016-01-07 16:38:43 +02001318 struct net *net = sock_net((struct sock *)tp);
1319
1320 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321}
1322
Eric Dumazetdf19a622009-08-28 23:48:54 -07001323static inline int keepalive_probes(const struct tcp_sock *tp)
1324{
Nikolay Borisov9bd68612016-01-07 16:38:44 +02001325 struct net *net = sock_net((struct sock *)tp);
1326
1327 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
Eric Dumazetdf19a622009-08-28 23:48:54 -07001328}
1329
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001330static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1331{
1332 const struct inet_connection_sock *icsk = &tp->inet_conn;
1333
Eric Dumazet70eabf02017-05-16 14:00:07 -07001334 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1335 tcp_jiffies32 - tp->rcv_tstamp);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001336}
1337
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001338static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339{
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02001340 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001341 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001343 if (fin_timeout < (rto << 2) - (rto >> 1))
1344 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
1346 return fin_timeout;
1347}
1348
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001349static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1350 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001352 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001353 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001354 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001355 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001356 /*
1357 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1358 * then following tcp messages have valid values. Ignore 0 value,
1359 * or else 'negative' tsval might forbid us to accept their packets.
1360 */
1361 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001362 return true;
1363 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001364}
1365
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001366static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1367 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001368{
1369 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001370 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
1372 /* RST segments are not recommended to carry timestamp,
1373 and, if they do, it is recommended to ignore PAWS because
1374 "their cleanup function should take precedence over timestamps."
1375 Certainly, it is mistake. It is necessary to understand the reasons
1376 of this constraint to relax it: if peer reboots, clock may go
1377 out-of-sync and half-open connections will not be reset.
1378 Actually, the problem would be not existing if all
1379 the implementations followed draft about maintaining clock
1380 via reboots. Linux-2.2 DOES NOT!
1381
1382 However, we can relax time bounds for RST segments to MSL.
1383 */
James Morris9d729f72007-03-04 16:12:44 -08001384 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001385 return false;
1386 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001389bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1390 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001391
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001392static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 /* See RFC 2012 */
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001395 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1396 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1397 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1398 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
1400
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001401/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001402static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001403{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001404 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001405}
1406
1407static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1408{
1409 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001410 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001411}
1412
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001413union tcp_md5_addr {
1414 struct in_addr a4;
1415#if IS_ENABLED(CONFIG_IPV6)
1416 struct in6_addr a6;
1417#endif
1418};
1419
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001420/* - key database */
1421struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001422 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001423 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001424 u8 family; /* AF_INET or AF_INET6 */
1425 union tcp_md5_addr addr;
Ivan Delalande67973182017-06-15 18:07:06 -07001426 u8 prefixlen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001427 u8 key[TCP_MD5SIG_MAXKEYLEN];
1428 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001429};
1430
1431/* - sock block */
1432struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001433 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001434 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001435};
1436
1437/* - pseudo header */
1438struct tcp4_pseudohdr {
1439 __be32 saddr;
1440 __be32 daddr;
1441 __u8 pad;
1442 __u8 protocol;
1443 __be16 len;
1444};
1445
1446struct tcp6_pseudohdr {
1447 struct in6_addr saddr;
1448 struct in6_addr daddr;
1449 __be32 len;
1450 __be32 protocol; /* including padding */
1451};
1452
1453union tcp_md5sum_block {
1454 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001455#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001456 struct tcp6_pseudohdr ip6;
1457#endif
1458};
1459
1460/* - pool: digest algorithm, hash description and scratch buffer */
1461struct tcp_md5sig_pool {
Herbert Xucf80e0e2016-01-24 21:20:23 +08001462 struct ahash_request *md5_req;
Eric Dumazet19689e32016-06-27 18:51:53 +02001463 void *scratch;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001464};
1465
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001466/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001467int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1468 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001469int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
Ivan Delalande67973182017-06-15 18:07:06 -07001470 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1471 gfp_t gfp);
Joe Perches5c9f3022013-09-23 11:33:32 -07001472int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
Ivan Delalande67973182017-06-15 18:07:06 -07001473 int family, u8 prefixlen);
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001474struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001475 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001476
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001477#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001478struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -07001479 const union tcp_md5_addr *addr,
1480 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001481#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001482#else
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001483static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001484 const union tcp_md5_addr *addr,
1485 int family)
1486{
1487 return NULL;
1488}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001489#define tcp_twsk_md5_key(twsk) NULL
1490#endif
1491
Joe Perches5c9f3022013-09-23 11:33:32 -07001492bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001493
Joe Perches5c9f3022013-09-23 11:33:32 -07001494struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001495static inline void tcp_put_md5sig_pool(void)
1496{
1497 local_bh_enable();
1498}
Eric Dumazet35790c02010-05-16 00:34:04 -07001499
Joe Perches5c9f3022013-09-23 11:33:32 -07001500int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1501 unsigned int header_len);
1502int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1503 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001504
Jerry Chu10467162012-08-31 12:29:11 +00001505/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001506void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1507 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1508 unsigned long *last_syn_loss);
1509void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -07001510 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1511 u16 try_exp);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001512struct tcp_fastopen_request {
1513 /* Fast Open cookie. Size 0 means a cookie request */
1514 struct tcp_fastopen_cookie cookie;
1515 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001516 size_t size;
1517 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001518};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001519void tcp_free_fastopen_req(struct tcp_sock *tp);
1520
Jerry Chu10467162012-08-31 12:29:11 +00001521extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1522int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Eric Dumazet61d2bca2016-02-01 21:03:07 -08001523void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
Eric Dumazet7c85af82015-09-24 17:16:05 -07001524struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1525 struct request_sock *req,
1526 struct tcp_fastopen_cookie *foc,
1527 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001528void tcp_fastopen_init_key_once(bool publish);
Wei Wang065263f2017-01-23 10:59:20 -08001529bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1530 struct tcp_fastopen_cookie *cookie);
Wei Wang19f6d3f2017-01-23 10:59:22 -08001531bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
Jerry Chu10467162012-08-31 12:29:11 +00001532#define TCP_FASTOPEN_KEY_LENGTH 16
1533
1534/* Fastopen key context */
1535struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001536 struct crypto_cipher *tfm;
1537 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1538 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001539};
1540
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001541extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
Wei Wang46c2fa32017-04-20 14:45:47 -07001542void tcp_fastopen_active_disable(struct sock *sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001543bool tcp_fastopen_active_should_disable(struct sock *sk);
1544void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1545void tcp_fastopen_active_timeout_reset(void);
1546
Francis Yan05b055e2016-11-27 23:07:13 -08001547/* Latencies incurred by various limits for a sender. They are
1548 * chronograph-like stats that are mutually exclusive.
1549 */
1550enum tcp_chrono {
1551 TCP_CHRONO_UNSPEC,
1552 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1553 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1554 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1555 __TCP_CHRONO_MAX,
1556};
1557
1558void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1559void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1560
David S. Millerfe067e82007-03-07 12:12:44 -08001561/* write queue abstraction */
1562static inline void tcp_write_queue_purge(struct sock *sk)
1563{
1564 struct sk_buff *skb;
1565
Francis Yan0f872302016-11-27 23:07:14 -08001566 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
David S. Millerfe067e82007-03-07 12:12:44 -08001567 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001568 sk_wmem_free_skb(sk, skb);
1569 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001570 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001571}
1572
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001573static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001574{
David S. Millercd07a8e2008-09-23 00:50:13 -07001575 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001576}
1577
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001578static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001579{
David S. Millercd07a8e2008-09-23 00:50:13 -07001580 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001581}
1582
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001583static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1584 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001585{
David S. Millercd07a8e2008-09-23 00:50:13 -07001586 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001587}
1588
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001589static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1590 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001591{
1592 return skb_queue_prev(&sk->sk_write_queue, skb);
1593}
1594
David S. Millerfe067e82007-03-07 12:12:44 -08001595#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001596 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001597
1598#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001599 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001600
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001601#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001602 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001603
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001604static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001605{
1606 return sk->sk_send_head;
1607}
1608
David S. Millercd07a8e2008-09-23 00:50:13 -07001609static inline bool tcp_skb_is_last(const struct sock *sk,
1610 const struct sk_buff *skb)
1611{
1612 return skb_queue_is_last(&sk->sk_write_queue, skb);
1613}
1614
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001615static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001616{
David S. Millercd07a8e2008-09-23 00:50:13 -07001617 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001618 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001619 else
1620 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001621}
1622
1623static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1624{
Francis Yan0f872302016-11-27 23:07:14 -08001625 if (sk->sk_send_head == skb_unlinked) {
David S. Millerfe067e82007-03-07 12:12:44 -08001626 sk->sk_send_head = NULL;
Francis Yan0f872302016-11-27 23:07:14 -08001627 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1628 }
Eric Dumazetbb1fcec2016-08-17 05:56:26 -07001629 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1630 tcp_sk(sk)->highest_sack = NULL;
David S. Millerfe067e82007-03-07 12:12:44 -08001631}
1632
1633static inline void tcp_init_send_head(struct sock *sk)
1634{
1635 sk->sk_send_head = NULL;
1636}
1637
1638static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1639{
1640 __skb_queue_tail(&sk->sk_write_queue, skb);
1641}
1642
1643static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1644{
1645 __tcp_add_write_queue_tail(sk, skb);
1646
1647 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001648 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001649 sk->sk_send_head = skb;
Francis Yan0f872302016-11-27 23:07:14 -08001650 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001651
1652 if (tcp_sk(sk)->highest_sack == NULL)
1653 tcp_sk(sk)->highest_sack = skb;
1654 }
David S. Millerfe067e82007-03-07 12:12:44 -08001655}
1656
1657static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1658{
1659 __skb_queue_head(&sk->sk_write_queue, skb);
1660}
1661
1662/* Insert buff after skb on the write queue of sk. */
1663static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1664 struct sk_buff *buff,
1665 struct sock *sk)
1666{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001667 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001668}
1669
David S. Miller43f59c82008-09-21 21:28:51 -07001670/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001671static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1672 struct sk_buff *skb,
1673 struct sock *sk)
1674{
David S. Miller43f59c82008-09-21 21:28:51 -07001675 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001676
1677 if (sk->sk_send_head == skb)
1678 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001679}
1680
1681static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1682{
1683 __skb_unlink(skb, &sk->sk_write_queue);
1684}
1685
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001686static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001687{
1688 return skb_queue_empty(&sk->sk_write_queue);
1689}
1690
Krishna Kumar12d50c42009-12-08 22:26:13 +00001691static inline void tcp_push_pending_frames(struct sock *sk)
1692{
1693 if (tcp_send_head(sk)) {
1694 struct tcp_sock *tp = tcp_sk(sk);
1695
1696 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1697 }
1698}
1699
Neal Cardwellecb97192012-02-27 17:52:52 -05001700/* Start sequence of the skb just after the highest skb with SACKed
1701 * bit, valid only if sacked_out > 0 or when the caller has ensured
1702 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001703 */
1704static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1705{
1706 if (!tp->sacked_out)
1707 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001708
1709 if (tp->highest_sack == NULL)
1710 return tp->snd_nxt;
1711
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001712 return TCP_SKB_CB(tp->highest_sack)->seq;
1713}
1714
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001715static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1716{
1717 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1718 tcp_write_queue_next(sk, skb);
1719}
1720
1721static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1722{
1723 return tcp_sk(sk)->highest_sack;
1724}
1725
1726static inline void tcp_highest_sack_reset(struct sock *sk)
1727{
1728 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1729}
1730
1731/* Called when old skb is about to be deleted (to be combined with new skb) */
1732static inline void tcp_highest_sack_combine(struct sock *sk,
1733 struct sk_buff *old,
1734 struct sk_buff *new)
1735{
1736 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1737 tcp_sk(sk)->highest_sack = new;
1738}
1739
Florian Westphalb1f0a0e2015-12-21 21:29:24 +01001740/* This helper checks if socket has IP_TRANSPARENT set */
1741static inline bool inet_sk_transparent(const struct sock *sk)
1742{
1743 switch (sk->sk_state) {
1744 case TCP_TIME_WAIT:
1745 return inet_twsk(sk)->tw_transparent;
1746 case TCP_NEW_SYN_RECV:
1747 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1748 }
1749 return inet_sk(sk)->transparent;
1750}
1751
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001752/* Determines whether this is a thin stream (which may suffer from
1753 * increased latency). Used to trigger latency-reducing mechanisms.
1754 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001755static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001756{
1757 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1758}
1759
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760/* /proc */
1761enum tcp_seq_states {
1762 TCP_SEQ_STATE_LISTENING,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764};
1765
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001766int tcp_seq_open(struct inode *inode, struct file *file);
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001769 char *name;
1770 sa_family_t family;
1771 const struct file_operations *seq_fops;
1772 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773};
1774
1775struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001776 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 sa_family_t family;
1778 enum tcp_seq_states state;
1779 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001780 int bucket, offset, sbucket, num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001781 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782};
1783
Joe Perches5c9f3022013-09-23 11:33:32 -07001784int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1785void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001787extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001788extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001789
Joe Perches5c9f3022013-09-23 11:33:32 -07001790void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001791
Eric Dumazet28be6e02013-10-18 10:36:17 -07001792struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001793 netdev_features_t features);
1794struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1795int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001796
Joe Perches5c9f3022013-09-23 11:33:32 -07001797void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001798
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001799static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1800{
Nikolay Borisov4979f2d2016-02-03 09:46:57 +02001801 struct net *net = sock_net((struct sock *)tp);
1802 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001803}
1804
1805static inline bool tcp_stream_memory_free(const struct sock *sk)
1806{
1807 const struct tcp_sock *tp = tcp_sk(sk);
1808 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1809
1810 return notsent_bytes < tcp_notsent_lowat(tp);
1811}
1812
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001813#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001814int tcp4_proc_init(void);
1815void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001816#endif
1817
Eric Dumazetea3bea32015-09-25 07:39:23 -07001818int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001819int tcp_conn_request(struct request_sock_ops *rsk_ops,
1820 const struct tcp_request_sock_ops *af_ops,
1821 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001822
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001823/* TCP af-specific functions */
1824struct tcp_sock_af_ops {
1825#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001826 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001827 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001828 int (*calc_md5_hash)(char *location,
1829 const struct tcp_md5sig_key *md5,
1830 const struct sock *sk,
1831 const struct sk_buff *skb);
1832 int (*md5_parse)(struct sock *sk,
Ivan Delalande8917a772017-06-15 18:07:07 -07001833 int optname,
Eric Dumazet39f8e582015-03-24 15:58:55 -07001834 char __user *optval,
1835 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001836#endif
1837};
1838
1839struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001840 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001841#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001842 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001843 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001844 int (*calc_md5_hash) (char *location,
1845 const struct tcp_md5sig_key *md5,
1846 const struct sock *sk,
1847 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001848#endif
Eric Dumazetb40cf182015-09-25 07:39:08 -07001849 void (*init_req)(struct request_sock *req,
1850 const struct sock *sk_listener,
Octavian Purdila16bea702014-06-25 17:09:53 +03001851 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001852#ifdef CONFIG_SYN_COOKIES
Eric Dumazet3f684b42015-09-29 07:42:49 -07001853 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001854 __u16 *mss);
1855#endif
Eric Dumazetf9646292015-09-29 07:42:50 -07001856 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
Soheil Hassas Yeganeh4396e462017-03-15 16:30:46 -04001857 const struct request_sock *req);
Eric Dumazet84b114b2017-05-05 06:56:54 -07001858 u32 (*init_seq)(const struct sk_buff *skb);
Eric Dumazet5d2ed052017-06-07 10:34:39 -07001859 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
Eric Dumazet0f935db2015-09-25 07:39:21 -07001860 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001861 struct flowi *fl, struct request_sock *req,
Eric Dumazetdc6ef6b2015-10-16 13:00:01 -07001862 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -07001863 enum tcp_synack_type synack_type);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001864};
1865
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001866#ifdef CONFIG_SYN_COOKIES
1867static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001868 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001869 __u16 *mss)
1870{
Eric Dumazet3f684b42015-09-29 07:42:49 -07001871 tcp_synq_overflow(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001872 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
Eric Dumazet3f684b42015-09-29 07:42:49 -07001873 return ops->cookie_init_seq(skb, mss);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001874}
1875#else
1876static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001877 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001878 __u16 *mss)
1879{
1880 return 0;
1881}
1882#endif
1883
Joe Perches5c9f3022013-09-23 11:33:32 -07001884int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001885
Joe Perches5c9f3022013-09-23 11:33:32 -07001886void tcp_v4_init(void);
1887void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001888
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001889/* tcp_recovery.c */
Eric Dumazet128eda82017-04-25 10:15:34 -07001890extern void tcp_rack_mark_lost(struct sock *sk);
Yuchung Cheng1d0833d2017-01-12 22:11:34 -08001891extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
Eric Dumazet9a568de2017-05-16 14:00:14 -07001892 u64 xmit_time);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -08001893extern void tcp_rack_reo_timeout(struct sock *sk);
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001894
Cong Wange25f8662014-10-15 14:33:21 -07001895/*
1896 * Save and compile IPv4 options, return a pointer to it
1897 */
Paolo Abeni91ed1e62017-08-03 18:07:06 +02001898static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1899 struct sk_buff *skb)
Cong Wange25f8662014-10-15 14:33:21 -07001900{
1901 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1902 struct ip_options_rcu *dopt = NULL;
1903
Cong Wang461b74c2014-10-15 14:33:22 -07001904 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07001905 int opt_size = sizeof(*dopt) + opt->optlen;
1906
1907 dopt = kmalloc(opt_size, GFP_ATOMIC);
Paolo Abeni91ed1e62017-08-03 18:07:06 +02001908 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
Cong Wange25f8662014-10-15 14:33:21 -07001909 kfree(dopt);
1910 dopt = NULL;
1911 }
1912 }
1913 return dopt;
1914}
1915
Eric Dumazet98781962015-02-03 18:31:53 -08001916/* locally generated TCP pure ACKs have skb->truesize == 2
1917 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1918 * This is much faster than dissecting the packet to find out.
1919 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1920 */
1921static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1922{
1923 return skb->truesize == 2;
1924}
1925
1926static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1927{
1928 skb->truesize = 2;
1929}
1930
Tom Herbert473bd232016-03-07 14:11:05 -08001931static inline int tcp_inq(struct sock *sk)
1932{
1933 struct tcp_sock *tp = tcp_sk(sk);
1934 int answ;
1935
1936 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1937 answ = 0;
1938 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1939 !tp->urg_data ||
1940 before(tp->urg_seq, tp->copied_seq) ||
1941 !before(tp->urg_seq, tp->rcv_nxt)) {
1942
1943 answ = tp->rcv_nxt - tp->copied_seq;
1944
1945 /* Subtract 1, if FIN was received */
1946 if (answ && sock_flag(sk, SOCK_DONE))
1947 answ--;
1948 } else {
1949 answ = tp->urg_seq - tp->copied_seq;
1950 }
1951
1952 return answ;
1953}
1954
Tom Herbert32035582016-08-28 14:43:18 -07001955int tcp_peek_len(struct socket *sock);
1956
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07001957static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1958{
1959 u16 segs_in;
1960
1961 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1962 tp->segs_in += segs_in;
1963 if (skb->len > tcp_hdrlen(skb))
1964 tp->data_segs_in += segs_in;
1965}
1966
Eric Dumazet9caad862016-04-01 08:52:20 -07001967/*
1968 * TCP listen path runs lockless.
1969 * We forced "struct sock" to be const qualified to make sure
1970 * we don't modify one of its field by mistake.
1971 * Here, we increment sk_drops which is an atomic_t, so we can safely
1972 * make sock writable again.
1973 */
1974static inline void tcp_listendrop(const struct sock *sk)
1975{
1976 atomic_inc(&((struct sock *)sk)->sk_drops);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001977 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
Eric Dumazet9caad862016-04-01 08:52:20 -07001978}
1979
Eric Dumazet218af592017-05-16 04:24:36 -07001980enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
1981
Dave Watson734942c2017-06-14 11:37:14 -07001982/*
1983 * Interface for adding Upper Level Protocols over TCP
1984 */
1985
1986#define TCP_ULP_NAME_MAX 16
1987#define TCP_ULP_MAX 128
1988#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
1989
1990struct tcp_ulp_ops {
1991 struct list_head list;
1992
1993 /* initialize ulp */
1994 int (*init)(struct sock *sk);
1995 /* cleanup ulp */
1996 void (*release)(struct sock *sk);
1997
1998 char name[TCP_ULP_NAME_MAX];
1999 struct module *owner;
2000};
2001int tcp_register_ulp(struct tcp_ulp_ops *type);
2002void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2003int tcp_set_ulp(struct sock *sk, const char *name);
2004void tcp_get_available_ulp(char *buf, size_t len);
2005void tcp_cleanup_ulp(struct sock *sk);
2006
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002007/* Call BPF_SOCK_OPS program that returns an int. If the return value
2008 * is < 0, then the BPF op failed (for example if the loaded BPF
2009 * program does not support the chosen operation or there is no BPF
2010 * program loaded).
2011 */
2012#ifdef CONFIG_BPF
2013static inline int tcp_call_bpf(struct sock *sk, int op)
2014{
2015 struct bpf_sock_ops_kern sock_ops;
2016 int ret;
2017
2018 if (sk_fullsock(sk))
2019 sock_owned_by_me(sk);
2020
2021 memset(&sock_ops, 0, sizeof(sock_ops));
2022 sock_ops.sk = sk;
2023 sock_ops.op = op;
2024
2025 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2026 if (ret == 0)
2027 ret = sock_ops.reply;
2028 else
2029 ret = -1;
2030 return ret;
2031}
2032#else
2033static inline int tcp_call_bpf(struct sock *sk, int op)
2034{
2035 return -EPERM;
2036}
2037#endif
2038
Lawrence Brakmo8550f322017-06-30 20:02:42 -07002039static inline u32 tcp_timeout_init(struct sock *sk)
2040{
2041 int timeout;
2042
2043 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT);
2044
2045 if (timeout <= 0)
2046 timeout = TCP_TIMEOUT_INIT;
2047 return timeout;
2048}
2049
Lawrence Brakmo13d3b1e2017-06-30 20:02:44 -07002050static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2051{
2052 int rwnd;
2053
2054 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT);
2055
2056 if (rwnd < 0)
2057 rwnd = 0;
2058 return rwnd;
2059}
Lawrence Brakmo91b5b212017-06-30 20:02:49 -07002060
2061static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2062{
2063 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1);
2064}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065#endif /* _TCP_H */