blob: d7047de952f0294b9eb3f9703eb74d8e573e44a5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#define FASTRETRANS_DEBUG 1
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/list.h>
24#include <linux/tcp.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050025#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
Herbert Xufb286bb2005-11-10 13:01:24 -080029#include <linux/skbuff.h>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080030#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000031#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080032#include <linux/ktime.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070033
34#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070035#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070036#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070038#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070042#include <net/tcp_states.h>
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -070043#include <net/inet_ecn.h>
Satoru SATOH0c266892009-05-04 11:11:01 -070044#include <net/dst.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/seq_file.h>
Glauber Costa180d8cd2011-12-11 21:47:02 +000047#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070049extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Eric Dumazetdd24c002008-11-25 21:17:14 -080051extern struct percpu_counter tcp_orphan_count;
Joe Perches5c9f3022013-09-23 11:33:32 -070052void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070055#define MAX_TCP_OPTION_SPACE 40
Eric Dumazetcc1b58c2019-06-15 17:31:03 -070056#define TCP_MIN_SND_MSS 48
57#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090059/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090061 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 */
63#define MAX_TCP_WINDOW 32767U
64
65/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66#define TCP_MIN_MSS 88U
67
John Heffner5d424d52006-03-20 17:53:41 -080068/* The least MTU to use for probing */
Fan Dudcd8fb82015-03-06 11:18:22 +080069#define TCP_BASE_MSS 1024
John Heffner5d424d52006-03-20 17:53:41 -080070
Fan Du05cbc0d2015-03-06 11:18:24 +080071/* probing interval, default to 10 minutes as per RFC4821 */
72#define TCP_PROBE_INTERVAL 600
73
Fan Du6b58e0a2015-03-06 11:18:23 +080074/* Specify interval when tcp mtu probing will stop */
75#define TCP_PROBE_THRESHOLD 8
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* After receiving this amount of duplicate ACKs fast retransmit starts. */
78#define TCP_FASTRETRANS_THRESH 3
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080/* Maximal number of ACKs sent quickly to accelerate slow-start. */
81#define TCP_MAX_QUICKACKS 16U
82
83/* urg_data states */
84#define TCP_URG_VALID 0x0100
85#define TCP_URG_NOTYET 0x0200
86#define TCP_URG_READ 0x0400
87
88#define TCP_RETR1 3 /*
89 * This is how many retries it does before it
90 * tries to figure out if the gateway is
91 * down. Minimal RFC value is 3; it corresponds
92 * to ~3sec-8min depending on RTO.
93 */
94
95#define TCP_RETR2 15 /*
96 * This should take at least
97 * 90 minutes to time out.
98 * RFC1122 says that the limit is 100 sec.
99 * 15 is ~13-30min depending on RTO.
100 */
101
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000102#define TCP_SYN_RETRIES 6 /* This is how many retries are done
103 * when active opening a connection.
104 * RFC1122 says the minimum retry MUST
105 * be at least 180secs. Nevertheless
106 * this value is corresponding to
107 * 63secs of retransmission with the
108 * current initial RTO.
109 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000111#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
112 * when passive opening a connection.
113 * This is corresponding to 31secs of
114 * retransmission with the current
115 * initial RTO.
116 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
119 * state, about 60 seconds */
120#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
121 /* BSD style FIN_WAIT2 deadlock breaker.
122 * It used to be 3min, new value is 60sec,
123 * to combine FIN-WAIT-2 timeout with
124 * TIME-WAIT timer.
125 */
126
127#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128#if HZ >= 100
129#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
130#define TCP_ATO_MIN ((unsigned)(HZ/25))
131#else
132#define TCP_DELACK_MIN 4U
133#define TCP_ATO_MIN 4U
134#endif
135#define TCP_RTO_MAX ((unsigned)(120*HZ))
136#define TCP_RTO_MIN ((unsigned)(HZ/5))
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000137#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000138#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
139 * used as a fallback RTO for the
140 * initial data transmission if no
141 * valid RTT sample has been acquired,
142 * most likely due to retrans in 3WHS.
143 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
146 * for local resources.
147 */
148
149#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
150#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
151#define TCP_KEEPALIVE_INTVL (75*HZ)
152
153#define MAX_TCP_KEEPIDLE 32767
154#define MAX_TCP_KEEPINTVL 32767
155#define MAX_TCP_KEEPCNT 127
156#define MAX_TCP_SYNCNT 127
157
158#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
161#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
162 * after this time. It should be equal
163 * (or greater than) TCP_TIMEWAIT_LEN
164 * to provide reliability equal to one
165 * provided by timewait state.
166 */
167#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
168 * timestamps. It must be less than
169 * minimal timewait lifetime.
170 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171/*
172 * TCP option
173 */
Kenjiro Nakayama105970f2014-10-20 18:15:50 +0900174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175#define TCPOPT_NOP 1 /* Padding */
176#define TCPOPT_EOL 0 /* End of options */
177#define TCPOPT_MSS 2 /* Segment size negotiating */
178#define TCPOPT_WINDOW 3 /* Window scaling */
179#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
180#define TCPOPT_SACK 5 /* SACK Block */
181#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800182#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Daniel Lee7f9b8382015-04-06 14:37:26 -0700183#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000184#define TCPOPT_EXP 254 /* Experimental */
185/* Magic number to be after the option value for sharing TCP
186 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 */
188#define TCPOPT_FASTOPEN_MAGIC 0xF989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
190/*
191 * TCP option lengths
192 */
193
194#define TCPOLEN_MSS 4
195#define TCPOLEN_WINDOW 3
196#define TCPOLEN_SACK_PERM 2
197#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800198#define TCPOLEN_MD5SIG 18
Daniel Lee7f9b8382015-04-06 14:37:26 -0700199#define TCPOLEN_FASTOPEN_BASE 2
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000200#define TCPOLEN_EXP_FASTOPEN_BASE 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202/* But this is what stacks really send out. */
203#define TCPOLEN_TSTAMP_ALIGNED 12
204#define TCPOLEN_WSCALE_ALIGNED 4
205#define TCPOLEN_SACKPERM_ALIGNED 4
206#define TCPOLEN_SACK_BASE 2
207#define TCPOLEN_SACK_BASE_ALIGNED 4
208#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800209#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700210#define TCPOLEN_MSS_ALIGNED 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212/* Flags in tp->nonagle */
213#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
214#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800215#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000217/* TCP thin-stream limits */
218#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
219
Jörg Thalheim21603fc2016-01-27 22:54:06 +0100220/* TCP initial congestion window as per rfc6928 */
David S. Miller442b9632011-02-02 17:05:11 -0800221#define TCP_INIT_CWND 10
222
Yuchung Chengcf60af02012-07-19 06:43:09 +0000223/* Bit Flags for sysctl_tcp_fastopen */
224#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000225#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000226#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af02012-07-19 06:43:09 +0000227
Jerry Chu10467162012-08-31 12:29:11 +0000228/* Accept SYN data w/o any cookie option */
229#define TFO_SERVER_COOKIE_NOT_REQD 0x200
230
231/* Force enable TFO on all listeners, i.e., not requiring the
Yuchung Chengcebc5cb2016-08-22 17:17:54 -0700232 * TCP_FASTOPEN socket option.
Jerry Chu10467162012-08-31 12:29:11 +0000233 */
234#define TFO_SERVER_WO_SOCKOPT1 0x400
Jerry Chu10467162012-08-31 12:29:11 +0000235
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700236extern struct inet_timewait_death_row tcp_death_row;
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239extern int sysctl_tcp_timestamps;
240extern int sysctl_tcp_window_scaling;
241extern int sysctl_tcp_sack;
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000242extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243extern int sysctl_tcp_retrans_collapse;
244extern int sysctl_tcp_stdurg;
245extern int sysctl_tcp_rfc1337;
246extern int sysctl_tcp_abort_on_overflow;
247extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248extern int sysctl_tcp_fack;
249extern int sysctl_tcp_reordering;
Eric Dumazetdca145f2014-10-27 21:45:24 -0700250extern int sysctl_tcp_max_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700252extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253extern int sysctl_tcp_wmem[3];
254extern int sysctl_tcp_rmem[3];
255extern int sysctl_tcp_app_win;
256extern int sysctl_tcp_adv_win_scale;
257extern int sysctl_tcp_tw_reuse;
258extern int sysctl_tcp_frto;
259extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261extern int sysctl_tcp_moderate_rcvbuf;
262extern int sysctl_tcp_tso_win_divisor;
Rick Jones15d99e02006-03-20 22:40:29 -0800263extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700264extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000265extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000266extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000267extern int sysctl_tcp_early_retrans;
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000268extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200269extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700270extern int sysctl_tcp_min_tso_segs;
Yuchung Chengf6722582015-10-16 21:57:42 -0700271extern int sysctl_tcp_min_rtt_wlen;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800272extern int sysctl_tcp_autocorking;
Neal Cardwell032ee422015-02-06 16:04:38 -0500273extern int sysctl_tcp_invalid_ratelimit;
Eric Dumazet43e122b2015-08-21 17:38:02 -0700274extern int sysctl_tcp_pacing_ss_ratio;
275extern int sysctl_tcp_pacing_ca_ratio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Eric Dumazet8d987e52010-11-09 23:24:26 +0000277extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800278extern struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279extern int tcp_memory_pressure;
280
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700281/* optimized version of sk_under_memory_pressure() for TCP sockets */
282static inline bool tcp_under_memory_pressure(const struct sock *sk)
283{
Johannes Weinerbaac50b2016-01-14 15:21:17 -0800284 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
285 mem_cgroup_under_socket_pressure(sk->sk_memcg))
Johannes Weinere8056052016-01-14 15:21:14 -0800286 return true;
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700287
288 return tcp_memory_pressure;
289}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
293 */
294
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000295static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800297 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800299#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
301/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000302static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
304 return seq3 - seq2 >= seq1 - seq2;
305}
306
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800307static inline bool tcp_out_of_memory(struct sock *sk)
308{
309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
311 return true;
312 return false;
313}
314
Eric Dumazeta6c5ea42015-05-15 12:39:26 -0700315void sk_forced_mem_schedule(struct sock *sk, int size);
316
David S. Millerad1af0f2010-08-25 02:27:49 -0700317static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700318{
David S. Millerad1af0f2010-08-25 02:27:49 -0700319 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
320 int orphans = percpu_counter_read_positive(ocp);
321
322 if (orphans << shift > sysctl_tcp_max_orphans) {
323 orphans = percpu_counter_sum_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans)
325 return true;
326 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700327 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700328}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
Joe Perches5c9f3022013-09-23 11:33:32 -0700330bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800331
Florian Westphala0f82f62009-04-19 09:43:48 +0000332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333extern struct proto tcp_prot;
334
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700335#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700336#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700337#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000338#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Joe Perches5c9f3022013-09-23 11:33:32 -0700340void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000341
Joe Perches5c9f3022013-09-23 11:33:32 -0700342void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Joe Perches5c9f3022013-09-23 11:33:32 -0700344void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Joe Perches5c9f3022013-09-23 11:33:32 -0700346void tcp_v4_early_demux(struct sk_buff *skb);
347int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
Joe Perches5c9f3022013-09-23 11:33:32 -0700349int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800350int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700351int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
352 int flags);
353void tcp_release_cb(struct sock *sk);
354void tcp_wfree(struct sk_buff *skb);
355void tcp_write_timer_handler(struct sock *sk);
356void tcp_delack_timer_handler(struct sock *sk);
357int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700358int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700359void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
360 const struct tcphdr *th, unsigned int len);
361void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700362int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
363void tcp_twsk_destructor(struct sock *sk);
364ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
365 struct pipe_inode_info *pipe, size_t len,
366 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800367
Eric Dumazet90cf17d2018-05-21 15:08:56 -0700368void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700369static inline void tcp_dec_quickack_mode(struct sock *sk,
370 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700372 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700373
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700374 if (icsk->icsk_ack.quick) {
375 if (pkts >= icsk->icsk_ack.quick) {
376 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700377 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700378 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700379 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700380 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
382}
383
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700384#define TCP_ECN_OK 1
385#define TCP_ECN_QUEUE_CWR 2
386#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000387#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700388
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000389enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 TCP_TW_SUCCESS = 0,
391 TCP_TW_RST = 1,
392 TCP_TW_ACK = 2,
393 TCP_TW_SYN = 3
394};
395
396
Joe Perches5c9f3022013-09-23 11:33:32 -0700397enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
398 struct sk_buff *skb,
399 const struct tcphdr *th);
400struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazet52452c52015-03-19 19:04:19 -0700401 struct request_sock *req, bool fastopen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700402int tcp_child_process(struct sock *parent, struct sock *child,
403 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400404void tcp_enter_loss(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700405void tcp_clear_retrans(struct tcp_sock *tp);
406void tcp_update_metrics(struct sock *sk);
407void tcp_init_metrics(struct sock *sk);
408void tcp_metrics_init(void);
409bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200410 bool paws_check, bool timestamps);
Joe Perches5c9f3022013-09-23 11:33:32 -0700411bool tcp_remember_stamp(struct sock *sk);
412bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
413void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
414void tcp_disable_fack(struct tcp_sock *tp);
415void tcp_close(struct sock *sk, long timeout);
416void tcp_init_sock(struct sock *sk);
417unsigned int tcp_poll(struct file *file, struct socket *sock,
418 struct poll_table_struct *wait);
419int tcp_getsockopt(struct sock *sk, int level, int optname,
420 char __user *optval, int __user *optlen);
421int tcp_setsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, unsigned int optlen);
423int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000424 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700425int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000426 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700427void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700428void tcp_syn_ack_timeout(const struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800429int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
430 int flags, int *addr_len);
Joe Perches5c9f3022013-09-23 11:33:32 -0700431void tcp_parse_options(const struct sk_buff *skb,
432 struct tcp_options_received *opt_rx,
433 int estab, struct tcp_fastopen_cookie *foc);
434const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/*
437 * TCP v4 functions exported for the inet6 API
438 */
439
Joe Perches5c9f3022013-09-23 11:33:32 -0700440void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400441void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet9cf74902016-02-02 19:31:12 -0800442void tcp_req_err(struct sock *sk, u32 seq, bool abort);
Joe Perches5c9f3022013-09-23 11:33:32 -0700443int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700444struct sock *tcp_create_openreq_child(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -0700445 struct request_sock *req,
446 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100447void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Eric Dumazet0c271712015-09-29 07:42:48 -0700448struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700449 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700450 struct dst_entry *dst,
451 struct request_sock *req_unhash,
452 bool *own_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700453int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
454int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
455int tcp_connect(struct sock *sk);
Eric Dumazetb3d05142016-04-13 22:05:39 -0700456enum tcp_synack_type {
457 TCP_SYNACK_NORMAL,
458 TCP_SYNACK_FASTOPEN,
459 TCP_SYNACK_COOKIE,
460};
Eric Dumazet5d062de2015-09-25 07:39:19 -0700461struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
Joe Perches5c9f3022013-09-23 11:33:32 -0700462 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700463 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -0700464 enum tcp_synack_type synack_type);
Joe Perches5c9f3022013-09-23 11:33:32 -0700465int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000467void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000468int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000469void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471/* From syncookies.c */
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700472struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
473 struct request_sock *req,
474 struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700475int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
476 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700477struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400478#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200479
Eric Dumazet63262312014-03-19 21:02:21 -0700480/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200481 * This counter is used both as a hash input and partially encoded into
482 * the cookie value. A cookie is only validated further if the delta
483 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700484 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200485 * the counter advances immediately after a cookie is generated).
486 */
Eric Dumazet264ea102015-05-14 14:26:56 -0700487#define MAX_SYNCOOKIE_AGE 2
488#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
489#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
490
491/* syncookies: remember time of last synqueue overflow
492 * But do not dirty this field too often (once per second is enough)
Eric Dumazet3f684b42015-09-29 07:42:49 -0700493 * It is racy as we do not hold a lock, but race is very minor.
Eric Dumazet264ea102015-05-14 14:26:56 -0700494 */
Eric Dumazet3f684b42015-09-29 07:42:49 -0700495static inline void tcp_synq_overflow(const struct sock *sk)
Eric Dumazet264ea102015-05-14 14:26:56 -0700496{
497 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
498 unsigned long now = jiffies;
499
500 if (time_after(now, last_overflow + HZ))
501 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
502}
503
504/* syncookies: no recent synqueue overflow on this listening socket? */
505static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
506{
507 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
508
509 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
510}
Florian Westphal8c27bd72013-09-20 22:32:55 +0200511
512static inline u32 tcp_cookie_time(void)
513{
Eric Dumazet63262312014-03-19 21:02:21 -0700514 u64 val = get_jiffies_64();
515
Eric Dumazet264ea102015-05-14 14:26:56 -0700516 do_div(val, TCP_SYNCOOKIE_PERIOD);
Eric Dumazet63262312014-03-19 21:02:21 -0700517 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200518}
519
Joe Perches5c9f3022013-09-23 11:33:32 -0700520u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
521 u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700522__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
Joe Perches5c9f3022013-09-23 11:33:32 -0700523__u32 cookie_init_timestamp(struct request_sock *req);
Florian Westphalf1673382014-11-03 17:35:02 +0100524bool cookie_timestamp_decode(struct tcp_options_received *opt);
525bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100526 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700527
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800528/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700529int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
530 u32 cookie);
531struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100532
Joe Perches5c9f3022013-09-23 11:33:32 -0700533u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
534 const struct tcphdr *th, u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700535__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400536#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537/* tcp_output.c */
538
Neal Cardwell1b3878c2016-09-19 23:39:18 -0400539u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
540 int min_tso_segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700541void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
542 int nonagle);
543bool tcp_may_send_now(struct sock *sk);
Eric Dumazet10d3be52016-04-21 10:55:23 -0700544int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
545int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700546void tcp_retransmit_timer(struct sock *sk);
547void tcp_xmit_retransmit_queue(struct sock *);
548void tcp_simple_retransmit(struct sock *);
549int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300550int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Joe Perches5c9f3022013-09-23 11:33:32 -0700552void tcp_send_probe0(struct sock *);
553void tcp_send_partial(struct sock *);
Eric Dumazete520af42015-05-06 14:26:25 -0700554int tcp_write_wakeup(struct sock *, int mib);
Joe Perches5c9f3022013-09-23 11:33:32 -0700555void tcp_send_fin(struct sock *sk);
556void tcp_send_active_reset(struct sock *sk, gfp_t priority);
557int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700558void tcp_push_one(struct sock *, unsigned int mss_now);
Yuchung Cheng57ec8822018-07-18 13:56:35 -0700559void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
Joe Perches5c9f3022013-09-23 11:33:32 -0700560void tcp_send_ack(struct sock *sk);
561void tcp_send_delayed_ack(struct sock *sk);
562void tcp_send_loss_probe(struct sock *sk);
563bool tcp_schedule_loss_probe(struct sock *sk);
Martin KaFai Laucfea5a62016-04-19 22:39:29 -0700564void tcp_skb_collapse_tstamp(struct sk_buff *skb,
565 const struct sk_buff *next_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
David S. Millera762a982005-07-05 15:18:51 -0700567/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700568void tcp_resume_early_retransmit(struct sock *sk);
569void tcp_rearm_rto(struct sock *sk);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700570void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700571void tcp_reset(struct sock *sk);
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -0700572void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
Eric Dumazete3e17b72016-02-06 11:16:28 -0800573void tcp_fin(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700576void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700577static inline void tcp_clear_xmit_timers(struct sock *sk)
578{
579 inet_csk_clear_xmit_timers(sk);
580}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
Joe Perches5c9f3022013-09-23 11:33:32 -0700582unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
583unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000584
585/* Bound MSS / TSO packet size with the half of the window */
586static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
587{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700588 int cutoff;
589
590 /* When peer uses tiny windows, there is no use in packetizing
591 * to sub-MSS pieces for the sake of SWS or making sure there
592 * are enough packets in the pipe for fast recovery.
593 *
594 * On the other hand, for extremely large MSS devices, handling
595 * smaller than MSS windows in this way does make sense.
596 */
Seymour, Shane M2631b792016-06-28 23:06:48 +0000597 if (tp->max_window > TCP_MSS_DEFAULT)
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700598 cutoff = (tp->max_window >> 1);
599 else
600 cutoff = tp->max_window;
601
602 if (cutoff && pktsize > cutoff)
603 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000604 else
605 return pktsize;
606}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300608/* tcp.c */
Eric Dumazet0df48c22015-04-28 15:28:17 -0700609void tcp_get_info(struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
611/* Read 'sendfile()'-style from a TCP socket */
Joe Perches5c9f3022013-09-23 11:33:32 -0700612int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
613 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Joe Perches5c9f3022013-09-23 11:33:32 -0700615void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Joe Perches5c9f3022013-09-23 11:33:32 -0700617int tcp_mtu_to_mss(struct sock *sk, int pmtu);
618int tcp_mss_to_mtu(struct sock *sk, int mss);
619void tcp_mtup_init(struct sock *sk);
620void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800621
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000622static inline void tcp_bound_rto(const struct sock *sk)
623{
624 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
625 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
626}
627
628static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
629{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800630 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000631}
632
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800633static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
636 ntohl(TCP_FLAG_ACK) |
637 snd_wnd);
638}
639
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800640static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
642 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
643}
644
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700645static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700647 struct tcp_sock *tp = tcp_sk(sk);
648
Yaogong Wang9f5afea2016-09-07 14:49:28 -0700649 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 tp->rcv_wnd &&
651 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
652 !tp->urg_data)
653 tcp_fast_path_on(tp);
654}
655
Satoru SATOH0c266892009-05-04 11:11:01 -0700656/* Compute the actual rto_min value */
657static inline u32 tcp_rto_min(struct sock *sk)
658{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400659 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700660 u32 rto_min = TCP_RTO_MIN;
661
662 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
663 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
664 return rto_min;
665}
666
Eric Dumazet740b0f12014-02-26 14:02:48 -0800667static inline u32 tcp_rto_min_us(struct sock *sk)
668{
669 return jiffies_to_usecs(tcp_rto_min(sk));
670}
671
Daniel Borkmann81164412015-01-05 23:57:48 +0100672static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
673{
674 return dst_metric_locked(dst, RTAX_CC_ALGO);
675}
676
Yuchung Chengf6722582015-10-16 21:57:42 -0700677/* Minimum RTT in usec. ~0 means not available. */
678static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
679{
Neal Cardwell64033892016-09-19 23:39:10 -0400680 return minmax_get(&tp->rtt_min);
Yuchung Chengf6722582015-10-16 21:57:42 -0700681}
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683/* Compute the actual receive window we are currently advertising.
684 * Rcv_nxt can be after the window if our peer push more data
685 * than the offered window.
686 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800687static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
689 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
690
691 if (win < 0)
692 win = 0;
693 return (u32) win;
694}
695
696/* Choose a new window, without checks for shrinking, and without
697 * scaling applied to the result. The caller does these things
698 * if necessary. This is a "raw" window selection.
699 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700700u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Pavel Emelyanovee995282012-04-19 03:40:39 +0000702void tcp_send_window_probe(struct sock *sk);
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704/* TCP timestamps are only 32-bits, this causes a slight
705 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800706 * of jiffies in the buffer control blocks below. We decided
707 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * casts with the following macro.
709 */
710#define tcp_time_stamp ((__u32)(jiffies))
711
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700712static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
713{
714 return skb->skb_mstamp.stamp_jiffies;
715}
716
717
Changli Gaoa3433f32010-06-12 14:01:43 +0000718#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
719
720#define TCPHDR_FIN 0x01
721#define TCPHDR_SYN 0x02
722#define TCPHDR_RST 0x04
723#define TCPHDR_PSH 0x08
724#define TCPHDR_ACK 0x10
725#define TCPHDR_URG 0x20
726#define TCPHDR_ECE 0x40
727#define TCPHDR_CWR 0x80
728
Daniel Borkmann49213552015-05-19 21:04:22 +0200729#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
730
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800731/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700732 * TCP per-packet control information to the transmission code.
733 * We also store the host-order sequence numbers in here too.
734 * This is 44 bytes if IPV6 is enabled.
735 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 */
737struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 __u32 seq; /* Starting sequence number */
739 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700740 union {
741 /* Note : tcp_tw_isn is used in input path only
742 * (isn chosen by tcp_timewait_state_process())
743 *
Eric Dumazetf69ad292015-06-11 09:15:18 -0700744 * tcp_gso_segs/size are used in write queue only,
745 * cf tcp_skb_pcount()/tcp_skb_mss()
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700746 */
747 __u32 tcp_tw_isn;
Eric Dumazetf69ad292015-06-11 09:15:18 -0700748 struct {
749 u16 tcp_gso_segs;
750 u16 tcp_gso_size;
751 };
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700752 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400753 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 __u8 sacked; /* State flags for SACK/FACK. */
756#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
757#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
758#define TCPCB_LOST 0x04 /* SKB is lost */
759#define TCPCB_TAGBITS 0x07 /* All tag bits */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400760#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400762#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
763 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000765 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
Soheil Hassas Yeganeh6b084922016-04-02 23:08:08 -0400766 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700767 eor:1, /* Is skb MSG_EOR marked? */
768 unused:6;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700770 union {
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700771 struct {
Yuchung Chengb9f64822016-09-19 23:39:14 -0400772 /* There is space for up to 24 bytes */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400773 __u32 in_flight:30,/* Bytes in flight at transmit */
774 is_app_limited:1, /* cwnd not fully used? */
775 unused:1;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400776 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
777 __u32 delivered;
778 /* start of send pipeline phase */
779 struct skb_mstamp first_tx_mstamp;
780 /* when we reached the "delivered" count */
781 struct skb_mstamp delivered_mstamp;
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700782 } tx; /* only used for outgoing skbs */
783 union {
784 struct inet_skb_parm h4;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700785#if IS_ENABLED(CONFIG_IPV6)
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700786 struct inet6_skb_parm h6;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700787#endif
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700788 } header; /* For incoming skbs */
789 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790};
791
792#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
793
Eric Dumazet870c3152014-10-17 09:17:20 -0700794
Eric Dumazet815afe12014-10-18 08:34:37 -0700795#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700796/* This is the variant of inet6_iif() that must be used by TCP,
797 * as TCP moves IP6CB into a different location in skb->cb[]
798 */
799static inline int tcp_v6_iif(const struct sk_buff *skb)
800{
David Aherna04a4802016-10-16 20:02:52 -0700801 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
David Ahern74b20582016-05-10 11:19:50 -0700802
803 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
Eric Dumazet870c3152014-10-17 09:17:20 -0700804}
Eric Dumazet815afe12014-10-18 08:34:37 -0700805#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700806
David Aherna04a4802016-10-16 20:02:52 -0700807/* TCP_SKB_CB reference means this can not be used from early demux */
808static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
809{
810#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
811 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
David Ahernda967862016-11-02 12:08:25 -0700812 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
David Aherna04a4802016-10-16 20:02:52 -0700813 return true;
814#endif
815 return false;
816}
817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818/* Due to TSO, an SKB can be composed of multiple actual
819 * packets. To keep these tracked properly, we use this.
820 */
821static inline int tcp_skb_pcount(const struct sk_buff *skb)
822{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700823 return TCP_SKB_CB(skb)->tcp_gso_segs;
824}
825
826static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
827{
828 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
829}
830
831static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
832{
833 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835
Eric Dumazetf69ad292015-06-11 09:15:18 -0700836/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837static inline int tcp_skb_mss(const struct sk_buff *skb)
838{
Eric Dumazetf69ad292015-06-11 09:15:18 -0700839 return TCP_SKB_CB(skb)->tcp_gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840}
841
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700842static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
843{
844 return likely(!TCP_SKB_CB(skb)->eor);
845}
846
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700847/* Events passed to congestion control interface */
848enum tcp_ca_event {
849 CA_EVENT_TX_START, /* first transmit when no packets in flight */
850 CA_EVENT_CWND_RESTART, /* congestion window restart */
851 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700852 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200853 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
854 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200855};
856
Florian Westphal98900922014-09-26 22:37:35 +0200857/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200858enum tcp_ca_ack_event_flags {
Florian Westphal98900922014-09-26 22:37:35 +0200859 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
860 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
861 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700862};
863
864/*
865 * Interface for adding new TCP congestion control handlers
866 */
867#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800868#define TCP_CA_MAX 128
869#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
870
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100871#define TCP_CA_UNSPEC 0
872
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200873/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700874#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200875/* Requires ECN/ECT set on all packets */
876#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700877
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700878union tcp_cc_info;
879
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700880struct ack_sample {
881 u32 pkts_acked;
882 s32 rtt_us;
Lawrence Brakmo6f094b92016-06-08 21:16:44 -0700883 u32 in_flight;
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700884};
885
Yuchung Chengb9f64822016-09-19 23:39:14 -0400886/* A rate sample measures the number of (original/retransmitted) data
887 * packets delivered "delivered" over an interval of time "interval_us".
888 * The tcp_rate.c code fills in the rate sample, and congestion
889 * control modules that define a cong_control function to run at the end
890 * of ACK processing can optionally chose to consult this sample when
891 * setting cwnd and pacing rate.
892 * A sample is invalid if "delivered" or "interval_us" is negative.
893 */
894struct rate_sample {
895 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
896 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
897 s32 delivered; /* number of packets delivered over interval */
898 long interval_us; /* time for tp->delivered to incr "delivered" */
899 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
900 int losses; /* number of packets marked lost upon ACK */
901 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
902 u32 prior_in_flight; /* in flight before this ACK */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400903 bool is_app_limited; /* is sample from packet with bubble in pipe? */
Yuchung Chengb9f64822016-09-19 23:39:14 -0400904 bool is_retrans; /* is sample from retransmission? */
905};
906
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700907struct tcp_congestion_ops {
908 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100909 u32 key;
910 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700911
912 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300913 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700914 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300915 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700916
917 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300918 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700919 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700920 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700921 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300922 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700923 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300924 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +0200925 /* call when ack arrives (optional) */
926 void (*in_ack_event)(struct sock *sk, u32 flags);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700927 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300928 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700929 /* hook for packet ack accounting (optional) */
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700930 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
Neal Cardwelled6e7262016-09-19 23:39:17 -0400931 /* suggest number of segments for each skb to transmit (optional) */
932 u32 (*tso_segs_goal)(struct sock *sk);
Yuchung Cheng77bfc172016-09-19 23:39:20 -0400933 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
934 u32 (*sndbuf_expand)(struct sock *sk);
Yuchung Chengc0402762016-09-19 23:39:21 -0400935 /* call when packets are delivered to update cwnd and pacing rate,
936 * after all the ca_state processing. (optional)
937 */
938 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300939 /* get info for inet_diag (optional) */
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700940 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
941 union tcp_cc_info *info);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700942
943 char name[TCP_CA_NAME_MAX];
944 struct module *owner;
945};
946
Joe Perches5c9f3022013-09-23 11:33:32 -0700947int tcp_register_congestion_control(struct tcp_congestion_ops *type);
948void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700949
Florian Westphal55d86942014-09-26 22:37:32 +0200950void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700951void tcp_init_congestion_control(struct sock *sk);
952void tcp_cleanup_congestion_control(struct sock *sk);
953int tcp_set_default_congestion_control(const char *name);
954void tcp_get_default_congestion_control(char *name);
955void tcp_get_available_congestion_control(char *buf, size_t len);
956void tcp_get_allowed_congestion_control(char *buf, size_t len);
957int tcp_set_allowed_congestion_control(char *allowed);
958int tcp_set_congestion_control(struct sock *sk, const char *name);
Neal Cardwelle73ebb02015-01-28 20:01:35 -0500959u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
960void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700961
Joe Perches5c9f3022013-09-23 11:33:32 -0700962u32 tcp_reno_ssthresh(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -0700963void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -0700964extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700965
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100966struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
Daniel Borkmannc3a8d942015-08-31 15:58:47 +0200967u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
Daniel Borkmannea697632015-01-05 23:57:47 +0100968#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100969char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +0100970#else
971static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
972{
973 return NULL;
974}
975#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100976
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200977static inline bool tcp_ca_needs_ecn(const struct sock *sk)
978{
979 const struct inet_connection_sock *icsk = inet_csk(sk);
980
981 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
982}
983
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300984static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700985{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300986 struct inet_connection_sock *icsk = inet_csk(sk);
987
988 if (icsk->icsk_ca_ops->set_state)
989 icsk->icsk_ca_ops->set_state(sk, ca_state);
990 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700991}
992
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300993static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700994{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300995 const struct inet_connection_sock *icsk = inet_csk(sk);
996
997 if (icsk->icsk_ca_ops->cwnd_event)
998 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700999}
1000
Yuchung Chengb9f64822016-09-19 23:39:14 -04001001/* From tcp_rate.c */
1002void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1003void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1004 struct rate_sample *rs);
1005void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
Yousuk Seunge74fe722017-12-07 13:41:34 -08001006 bool is_sack_reneg, struct skb_mstamp *now, struct rate_sample *rs);
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001007void tcp_rate_check_app_limited(struct sock *sk);
Yuchung Chengb9f64822016-09-19 23:39:14 -04001008
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001009/* These functions determine how the current flow behaves in respect of SACK
1010 * handling. SACK is negotiated with the peer, and therefore it can vary
1011 * between different flows.
1012 *
1013 * tcp_is_sack - SACK enabled
1014 * tcp_is_reno - No SACK
1015 * tcp_is_fack - FACK enabled, implies SACK enabled
1016 */
1017static inline int tcp_is_sack(const struct tcp_sock *tp)
1018{
1019 return tp->rx_opt.sack_ok;
1020}
1021
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001022static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001023{
1024 return !tcp_is_sack(tp);
1025}
1026
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001027static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001028{
Vijay Subramanianab562222011-12-20 13:23:24 +00001029 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001030}
1031
1032static inline void tcp_enable_fack(struct tcp_sock *tp)
1033{
Vijay Subramanianab562222011-12-20 13:23:24 +00001034 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001035}
1036
Yuchung Chengeed530b2012-05-02 13:30:03 +00001037/* TCP early-retransmit (ER) is similar to but more conservative than
1038 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
1039 */
1040static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1041{
Nikolay Borisov1043e252016-02-03 09:46:52 +02001042 struct net *net = sock_net((struct sock *)tp);
1043
Yuchung Chengeed530b2012-05-02 13:30:03 +00001044 tp->do_early_retrans = sysctl_tcp_early_retrans &&
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +00001045 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
Nikolay Borisov1043e252016-02-03 09:46:52 +02001046 net->ipv4.sysctl_tcp_reordering == 3;
Yuchung Chengeed530b2012-05-02 13:30:03 +00001047}
1048
1049static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1050{
1051 tp->do_early_retrans = 0;
1052}
1053
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001054static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1055{
1056 return tp->sacked_out + tp->lost_out;
1057}
1058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059/* This determines how many packets are "in the network" to the best
1060 * of our knowledge. In many cases it is conservative, but where
1061 * detailed information is available from the receiver (via SACK
1062 * blocks etc.) we can make more aggressive calculations.
1063 *
1064 * Use this for decisions involving congestion control, use just
1065 * tp->packets_out to determine if the send queue is empty or not.
1066 *
1067 * Read this equation as:
1068 *
1069 * "Packets sent once on transmission queue" MINUS
1070 * "Packets left network, but not honestly ACKed yet" PLUS
1071 * "Packets fast retransmitted"
1072 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001073static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001075 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076}
1077
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001078#define TCP_INFINITE_SSTHRESH 0x7fffffff
1079
Yuchung Cheng071d5082015-07-09 13:16:29 -07001080static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1081{
Yuchung Cheng76174002015-07-09 13:16:30 -07001082 return tp->snd_cwnd < tp->snd_ssthresh;
Yuchung Cheng071d5082015-07-09 13:16:29 -07001083}
1084
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001085static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1086{
1087 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1088}
1089
Yuchung Cheng684bad12012-09-02 17:38:04 +00001090static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1091{
1092 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1093 (1 << inet_csk(sk)->icsk_ca_state);
1094}
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +00001097 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 * ssthresh.
1099 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001100static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001102 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001103
Yuchung Cheng684bad12012-09-02 17:38:04 +00001104 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 return tp->snd_ssthresh;
1106 else
1107 return max(tp->snd_ssthresh,
1108 ((tp->snd_cwnd >> 1) +
1109 (tp->snd_cwnd >> 2)));
1110}
1111
Ilpo Järvinenb9c45952007-07-27 16:36:17 +03001112/* Use define here intentionally to get WARN_ON location shown at the caller */
1113#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Christoph Paasch5ee2c942014-07-14 16:58:32 +02001115void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001116__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
Neal Cardwell6b5a5c02011-11-21 17:15:14 +00001118/* The maximum number of MSS of available cwnd for which TSO defers
1119 * sending if not using sysctl_tcp_tso_win_divisor.
1120 */
1121static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1122{
1123 return 3;
1124}
1125
Ilpo Järvinen90840de2007-12-31 04:48:41 -08001126/* Returns end sequence number of the receiver's advertised window */
1127static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1128{
1129 return tp->snd_una + tp->snd_wnd;
1130}
Eric Dumazete114a712014-04-30 11:58:13 -07001131
1132/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1133 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001134 * it was fully used previously. And that's exactly what we do in
1135 * congestion avoidance mode. But in slow start we allow cwnd to grow
1136 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001137 * Example :
1138 * cwnd is 10 (IW10), but application sends 9 frames.
1139 * We allow cwnd to reach 18 when all frames are ACKed.
1140 * This check is safe because it's as aggressive as slow start which already
1141 * risks 100% overshoot. The advantage is that we discourage application to
1142 * either send more filler packets or data to artificially blow up the cwnd
1143 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001144 */
Eric Dumazet24901552014-05-02 21:18:05 -07001145static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001146{
1147 const struct tcp_sock *tp = tcp_sk(sk);
1148
Neal Cardwellca8a2262014-05-22 10:41:08 -04001149 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
Yuchung Cheng071d5082015-07-09 13:16:29 -07001150 if (tcp_in_slow_start(tp))
Neal Cardwellca8a2262014-05-22 10:41:08 -04001151 return tp->snd_cwnd < 2 * tp->max_packets_out;
1152
1153 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001154}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001155
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001156/* Something is really bad, we could not queue an additional packet,
1157 * because qdisc is full or receiver sent a 0 window.
1158 * We do not want to add fuel to the fire, or abort too early,
1159 * so make sure the timer we arm now is at least 200ms in the future,
1160 * regardless of current icsk_rto value (as it could be ~2ms)
1161 */
1162static inline unsigned long tcp_probe0_base(const struct sock *sk)
1163{
1164 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1165}
1166
1167/* Variant of inet_csk_rto_backoff() used for zero window probes */
1168static inline unsigned long tcp_probe0_when(const struct sock *sk,
1169 unsigned long max_when)
1170{
1171 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1172
1173 return (unsigned long)min_t(u64, when, max_when);
1174}
1175
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001176static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001178 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001179 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001180 tcp_probe0_base(sk), TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001183static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
1185 tp->snd_wl1 = seq;
1186}
1187
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001188static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
1190 tp->snd_wl1 = seq;
1191}
1192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193/*
1194 * Calculate(/check) TCP checksum
1195 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001196static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1197 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
1199 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1200}
1201
Al Virob51655b2006-11-14 21:40:42 -08001202static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
Herbert Xufb286bb2005-11-10 13:01:24 -08001204 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001207static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
Herbert Xu60476372007-04-09 11:59:39 -07001209 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 __tcp_checksum_complete(skb);
1211}
1212
1213/* Prequeue for VJ style copy to user, combined with checksumming. */
1214
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001215static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
1217 tp->ucopy.task = NULL;
1218 tp->ucopy.len = 0;
1219 tp->ucopy.memory = 0;
1220 skb_queue_head_init(&tp->ucopy.prequeue);
1221}
1222
Joe Perches5c9f3022013-09-23 11:33:32 -07001223bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc9c33212016-08-27 07:37:54 -07001224bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
Eric Dumazetac6e7802016-11-10 13:12:35 -08001225int tcp_filter(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227#undef STATE_TRACE
1228
1229#ifdef STATE_TRACE
1230static const char *statename[]={
1231 "Unused","Established","Syn Sent","Syn Recv",
1232 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1233 "Close Wait","Last ACK","Listen","Closing"
1234};
1235#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001236void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Joe Perches5c9f3022013-09-23 11:33:32 -07001238void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09001240int tcp_abort(struct sock *sk, int err);
1241
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001242static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
1244 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 rx_opt->num_sacks = 0;
1246}
1247
Joe Perches5c9f3022013-09-23 11:33:32 -07001248u32 tcp_default_init_rwnd(u32 mss);
Eric Dumazet6f021c62015-08-21 12:30:00 -07001249void tcp_cwnd_restart(struct sock *sk, s32 delta);
1250
1251static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1252{
1253 struct tcp_sock *tp = tcp_sk(sk);
1254 s32 delta;
1255
1256 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1257 return;
1258 delta = tcp_time_stamp - tp->lsndtime;
1259 if (delta > inet_csk(sk)->icsk_rto)
1260 tcp_cwnd_restart(sk, delta);
1261}
Yuchung Cheng85f16522013-06-11 15:35:32 -07001262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001264void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1265 __u32 *window_clamp, int wscale_ok,
1266 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
1268static inline int tcp_win_from_space(int space)
1269{
Gao Feng694f2192017-03-24 07:05:12 +08001270 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1271
1272 return tcp_adv_win_scale <= 0 ?
1273 (space>>(-tcp_adv_win_scale)) :
1274 space - (space>>tcp_adv_win_scale);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275}
1276
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001277/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278static inline int tcp_space(const struct sock *sk)
1279{
1280 return tcp_win_from_space(sk->sk_rcvbuf -
1281 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001282}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
1284static inline int tcp_full_space(const struct sock *sk)
1285{
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001286 return tcp_win_from_space(sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001289extern void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -07001290 const struct sock *sk_listener,
1291 const struct dst_entry *dst);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001292
Joe Perches5c9f3022013-09-23 11:33:32 -07001293void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1296{
Nikolay Borisovb840d152016-01-07 16:38:45 +02001297 struct net *net = sock_net((struct sock *)tp);
1298
1299 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300}
1301
1302static inline int keepalive_time_when(const struct tcp_sock *tp)
1303{
Nikolay Borisov13b287e2016-01-07 16:38:43 +02001304 struct net *net = sock_net((struct sock *)tp);
1305
1306 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
1308
Eric Dumazetdf19a622009-08-28 23:48:54 -07001309static inline int keepalive_probes(const struct tcp_sock *tp)
1310{
Nikolay Borisov9bd68612016-01-07 16:38:44 +02001311 struct net *net = sock_net((struct sock *)tp);
1312
1313 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
Eric Dumazetdf19a622009-08-28 23:48:54 -07001314}
1315
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001316static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1317{
1318 const struct inet_connection_sock *icsk = &tp->inet_conn;
1319
1320 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1321 tcp_time_stamp - tp->rcv_tstamp);
1322}
1323
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001324static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02001326 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001327 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001329 if (fin_timeout < (rto << 2) - (rto >> 1))
1330 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332 return fin_timeout;
1333}
1334
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001335static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1336 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001338 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001339 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001340 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001341 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001342 /*
1343 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1344 * then following tcp messages have valid values. Ignore 0 value,
1345 * or else 'negative' tsval might forbid us to accept their packets.
1346 */
1347 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001348 return true;
1349 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001350}
1351
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001352static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1353 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001354{
1355 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001356 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 /* RST segments are not recommended to carry timestamp,
1359 and, if they do, it is recommended to ignore PAWS because
1360 "their cleanup function should take precedence over timestamps."
1361 Certainly, it is mistake. It is necessary to understand the reasons
1362 of this constraint to relax it: if peer reboots, clock may go
1363 out-of-sync and half-open connections will not be reset.
1364 Actually, the problem would be not existing if all
1365 the implementations followed draft about maintaining clock
1366 via reboots. Linux-2.2 DOES NOT!
1367
1368 However, we can relax time bounds for RST segments to MSL.
1369 */
James Morris9d729f72007-03-04 16:12:44 -08001370 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001371 return false;
1372 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373}
1374
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001375bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1376 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001377
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001378static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
1380 /* See RFC 2012 */
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001381 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1382 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1383 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1384 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385}
1386
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001387/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001388static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001389{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001390 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001391}
1392
1393static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1394{
1395 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001396 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001397}
1398
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001399union tcp_md5_addr {
1400 struct in_addr a4;
1401#if IS_ENABLED(CONFIG_IPV6)
1402 struct in6_addr a6;
1403#endif
1404};
1405
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001406/* - key database */
1407struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001408 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001409 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001410 u8 family; /* AF_INET or AF_INET6 */
1411 union tcp_md5_addr addr;
1412 u8 key[TCP_MD5SIG_MAXKEYLEN];
1413 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001414};
1415
1416/* - sock block */
1417struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001418 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001419 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001420};
1421
1422/* - pseudo header */
1423struct tcp4_pseudohdr {
1424 __be32 saddr;
1425 __be32 daddr;
1426 __u8 pad;
1427 __u8 protocol;
1428 __be16 len;
1429};
1430
1431struct tcp6_pseudohdr {
1432 struct in6_addr saddr;
1433 struct in6_addr daddr;
1434 __be32 len;
1435 __be32 protocol; /* including padding */
1436};
1437
1438union tcp_md5sum_block {
1439 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001440#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001441 struct tcp6_pseudohdr ip6;
1442#endif
1443};
1444
1445/* - pool: digest algorithm, hash description and scratch buffer */
1446struct tcp_md5sig_pool {
Herbert Xucf80e0e2016-01-24 21:20:23 +08001447 struct ahash_request *md5_req;
Eric Dumazet19689e32016-06-27 18:51:53 +02001448 void *scratch;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001449};
1450
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001451/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001452int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1453 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001454int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1455 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1456int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1457 int family);
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001458struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001459 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001460
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001461#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001462struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -07001463 const union tcp_md5_addr *addr,
1464 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001465#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001466#else
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001467static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001468 const union tcp_md5_addr *addr,
1469 int family)
1470{
1471 return NULL;
1472}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001473#define tcp_twsk_md5_key(twsk) NULL
1474#endif
1475
Joe Perches5c9f3022013-09-23 11:33:32 -07001476bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001477
Joe Perches5c9f3022013-09-23 11:33:32 -07001478struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001479static inline void tcp_put_md5sig_pool(void)
1480{
1481 local_bh_enable();
1482}
Eric Dumazet35790c02010-05-16 00:34:04 -07001483
Joe Perches5c9f3022013-09-23 11:33:32 -07001484int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1485 unsigned int header_len);
1486int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1487 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001488
Jerry Chu10467162012-08-31 12:29:11 +00001489/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001490void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1491 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1492 unsigned long *last_syn_loss);
1493void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -07001494 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1495 u16 try_exp);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001496struct tcp_fastopen_request {
1497 /* Fast Open cookie. Size 0 means a cookie request */
1498 struct tcp_fastopen_cookie cookie;
1499 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001500 size_t size;
1501 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001502};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001503void tcp_free_fastopen_req(struct tcp_sock *tp);
1504
Jerry Chu10467162012-08-31 12:29:11 +00001505extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1506int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Eric Dumazet61d2bca2016-02-01 21:03:07 -08001507void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
Eric Dumazet7c85af82015-09-24 17:16:05 -07001508struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1509 struct request_sock *req,
1510 struct tcp_fastopen_cookie *foc,
1511 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001512void tcp_fastopen_init_key_once(bool publish);
Jerry Chu10467162012-08-31 12:29:11 +00001513#define TCP_FASTOPEN_KEY_LENGTH 16
1514
1515/* Fastopen key context */
1516struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001517 struct crypto_cipher *tfm;
1518 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1519 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001520};
1521
David S. Millerfe067e82007-03-07 12:12:44 -08001522/* write queue abstraction */
1523static inline void tcp_write_queue_purge(struct sock *sk)
1524{
1525 struct sk_buff *skb;
1526
1527 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001528 sk_wmem_free_skb(sk, skb);
1529 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001530 tcp_clear_all_retrans_hints(tcp_sk(sk));
Eric Dumazet1f52cfe2019-02-15 13:36:20 -08001531 inet_csk(sk)->icsk_backoff = 0;
David S. Millerfe067e82007-03-07 12:12:44 -08001532}
1533
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001534static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001535{
David S. Millercd07a8e2008-09-23 00:50:13 -07001536 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001537}
1538
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001539static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001540{
David S. Millercd07a8e2008-09-23 00:50:13 -07001541 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001542}
1543
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001544static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1545 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001546{
David S. Millercd07a8e2008-09-23 00:50:13 -07001547 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001548}
1549
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001550static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1551 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001552{
1553 return skb_queue_prev(&sk->sk_write_queue, skb);
1554}
1555
David S. Millerfe067e82007-03-07 12:12:44 -08001556#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001557 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001558
1559#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001560 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001561
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001562#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001563 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001564
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001565static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001566{
1567 return sk->sk_send_head;
1568}
1569
David S. Millercd07a8e2008-09-23 00:50:13 -07001570static inline bool tcp_skb_is_last(const struct sock *sk,
1571 const struct sk_buff *skb)
1572{
1573 return skb_queue_is_last(&sk->sk_write_queue, skb);
1574}
1575
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001576static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001577{
David S. Millercd07a8e2008-09-23 00:50:13 -07001578 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001579 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001580 else
1581 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001582}
1583
1584static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1585{
1586 if (sk->sk_send_head == skb_unlinked)
1587 sk->sk_send_head = NULL;
Eric Dumazetbb1fcec2016-08-17 05:56:26 -07001588 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1589 tcp_sk(sk)->highest_sack = NULL;
David S. Millerfe067e82007-03-07 12:12:44 -08001590}
1591
1592static inline void tcp_init_send_head(struct sock *sk)
1593{
1594 sk->sk_send_head = NULL;
1595}
1596
1597static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1598{
1599 __skb_queue_tail(&sk->sk_write_queue, skb);
1600}
1601
1602static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1603{
1604 __tcp_add_write_queue_tail(sk, skb);
1605
1606 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001607 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001608 sk->sk_send_head = skb;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001609
1610 if (tcp_sk(sk)->highest_sack == NULL)
1611 tcp_sk(sk)->highest_sack = skb;
1612 }
David S. Millerfe067e82007-03-07 12:12:44 -08001613}
1614
1615static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1616{
1617 __skb_queue_head(&sk->sk_write_queue, skb);
1618}
1619
1620/* Insert buff after skb on the write queue of sk. */
1621static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1622 struct sk_buff *buff,
1623 struct sock *sk)
1624{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001625 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001626}
1627
David S. Miller43f59c82008-09-21 21:28:51 -07001628/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001629static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1630 struct sk_buff *skb,
1631 struct sock *sk)
1632{
David S. Miller43f59c82008-09-21 21:28:51 -07001633 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001634
1635 if (sk->sk_send_head == skb)
1636 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001637}
1638
1639static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1640{
1641 __skb_unlink(skb, &sk->sk_write_queue);
1642}
1643
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001644static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001645{
1646 return skb_queue_empty(&sk->sk_write_queue);
1647}
1648
Krishna Kumar12d50c42009-12-08 22:26:13 +00001649static inline void tcp_push_pending_frames(struct sock *sk)
1650{
1651 if (tcp_send_head(sk)) {
1652 struct tcp_sock *tp = tcp_sk(sk);
1653
1654 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1655 }
1656}
1657
Neal Cardwellecb97192012-02-27 17:52:52 -05001658/* Start sequence of the skb just after the highest skb with SACKed
1659 * bit, valid only if sacked_out > 0 or when the caller has ensured
1660 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001661 */
1662static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1663{
1664 if (!tp->sacked_out)
1665 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001666
1667 if (tp->highest_sack == NULL)
1668 return tp->snd_nxt;
1669
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001670 return TCP_SKB_CB(tp->highest_sack)->seq;
1671}
1672
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001673static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1674{
1675 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1676 tcp_write_queue_next(sk, skb);
1677}
1678
1679static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1680{
1681 return tcp_sk(sk)->highest_sack;
1682}
1683
1684static inline void tcp_highest_sack_reset(struct sock *sk)
1685{
1686 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1687}
1688
Eric Dumazete12c42c2017-10-30 23:08:20 -07001689/* Called when old skb is about to be deleted and replaced by new skb */
1690static inline void tcp_highest_sack_replace(struct sock *sk,
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001691 struct sk_buff *old,
1692 struct sk_buff *new)
1693{
Eric Dumazete12c42c2017-10-30 23:08:20 -07001694 if (old == tcp_highest_sack(sk))
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001695 tcp_sk(sk)->highest_sack = new;
1696}
1697
Florian Westphalb1f0a0e2015-12-21 21:29:24 +01001698/* This helper checks if socket has IP_TRANSPARENT set */
1699static inline bool inet_sk_transparent(const struct sock *sk)
1700{
1701 switch (sk->sk_state) {
1702 case TCP_TIME_WAIT:
1703 return inet_twsk(sk)->tw_transparent;
1704 case TCP_NEW_SYN_RECV:
1705 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1706 }
1707 return inet_sk(sk)->transparent;
1708}
1709
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001710/* Determines whether this is a thin stream (which may suffer from
1711 * increased latency). Used to trigger latency-reducing mechanisms.
1712 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001713static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001714{
1715 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1716}
1717
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718/* /proc */
1719enum tcp_seq_states {
1720 TCP_SEQ_STATE_LISTENING,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722};
1723
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001724int tcp_seq_open(struct inode *inode, struct file *file);
1725
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001727 char *name;
1728 sa_family_t family;
1729 const struct file_operations *seq_fops;
1730 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731};
1732
1733struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001734 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 sa_family_t family;
1736 enum tcp_seq_states state;
1737 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001738 int bucket, offset, sbucket, num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001739 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740};
1741
Joe Perches5c9f3022013-09-23 11:33:32 -07001742int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1743void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001745extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001746extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001747
Joe Perches5c9f3022013-09-23 11:33:32 -07001748void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001749
Eric Dumazet28be6e02013-10-18 10:36:17 -07001750struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001751 netdev_features_t features);
1752struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1753int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001754
Joe Perches5c9f3022013-09-23 11:33:32 -07001755void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001756
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001757static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1758{
Nikolay Borisov4979f2d2016-02-03 09:46:57 +02001759 struct net *net = sock_net((struct sock *)tp);
1760 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001761}
1762
1763static inline bool tcp_stream_memory_free(const struct sock *sk)
1764{
1765 const struct tcp_sock *tp = tcp_sk(sk);
1766 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1767
1768 return notsent_bytes < tcp_notsent_lowat(tp);
1769}
1770
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001771#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001772int tcp4_proc_init(void);
1773void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001774#endif
1775
Eric Dumazetea3bea32015-09-25 07:39:23 -07001776int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001777int tcp_conn_request(struct request_sock_ops *rsk_ops,
1778 const struct tcp_request_sock_ops *af_ops,
1779 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001780
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001781/* TCP af-specific functions */
1782struct tcp_sock_af_ops {
1783#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001784 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001785 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001786 int (*calc_md5_hash)(char *location,
1787 const struct tcp_md5sig_key *md5,
1788 const struct sock *sk,
1789 const struct sk_buff *skb);
1790 int (*md5_parse)(struct sock *sk,
1791 char __user *optval,
1792 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001793#endif
1794};
1795
1796struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001797 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001798#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001799 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001800 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001801 int (*calc_md5_hash) (char *location,
1802 const struct tcp_md5sig_key *md5,
1803 const struct sock *sk,
1804 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001805#endif
Eric Dumazetb40cf182015-09-25 07:39:08 -07001806 void (*init_req)(struct request_sock *req,
1807 const struct sock *sk_listener,
Octavian Purdila16bea702014-06-25 17:09:53 +03001808 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001809#ifdef CONFIG_SYN_COOKIES
Eric Dumazet3f684b42015-09-29 07:42:49 -07001810 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001811 __u16 *mss);
1812#endif
Eric Dumazetf9646292015-09-29 07:42:50 -07001813 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
Octavian Purdilad94e0412014-06-25 17:09:55 +03001814 const struct request_sock *req,
1815 bool *strict);
Octavian Purdila936b8bd2014-06-25 17:09:57 +03001816 __u32 (*init_seq)(const struct sk_buff *skb);
Eric Dumazet0f935dbe2015-09-25 07:39:21 -07001817 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001818 struct flowi *fl, struct request_sock *req,
Eric Dumazetdc6ef6b2015-10-16 13:00:01 -07001819 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -07001820 enum tcp_synack_type synack_type);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001821};
1822
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001823#ifdef CONFIG_SYN_COOKIES
1824static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001825 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001826 __u16 *mss)
1827{
Eric Dumazet3f684b42015-09-29 07:42:49 -07001828 tcp_synq_overflow(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001829 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
Eric Dumazet3f684b42015-09-29 07:42:49 -07001830 return ops->cookie_init_seq(skb, mss);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001831}
1832#else
1833static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001834 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001835 __u16 *mss)
1836{
1837 return 0;
1838}
1839#endif
1840
Joe Perches5c9f3022013-09-23 11:33:32 -07001841int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001842
Joe Perches5c9f3022013-09-23 11:33:32 -07001843void tcp_v4_init(void);
1844void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001845
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001846/* tcp_recovery.c */
1847
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -07001848/* Flags to enable various loss recovery features. See below */
1849extern int sysctl_tcp_recovery;
1850
1851/* Use TCP RACK to detect (some) tail and retransmit losses */
1852#define TCP_RACK_LOST_RETRANS 0x1
1853
1854extern int tcp_rack_mark_lost(struct sock *sk);
1855
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001856extern void tcp_rack_advance(struct tcp_sock *tp,
1857 const struct skb_mstamp *xmit_time, u8 sacked);
1858
Cong Wange25f8662014-10-15 14:33:21 -07001859/*
1860 * Save and compile IPv4 options, return a pointer to it
1861 */
1862static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1863{
1864 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1865 struct ip_options_rcu *dopt = NULL;
1866
Cong Wang461b74c2014-10-15 14:33:22 -07001867 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07001868 int opt_size = sizeof(*dopt) + opt->optlen;
1869
1870 dopt = kmalloc(opt_size, GFP_ATOMIC);
1871 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1872 kfree(dopt);
1873 dopt = NULL;
1874 }
1875 }
1876 return dopt;
1877}
1878
Eric Dumazet98781962015-02-03 18:31:53 -08001879/* locally generated TCP pure ACKs have skb->truesize == 2
1880 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1881 * This is much faster than dissecting the packet to find out.
1882 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1883 */
1884static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1885{
1886 return skb->truesize == 2;
1887}
1888
1889static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1890{
1891 skb->truesize = 2;
1892}
1893
Tom Herbert473bd232016-03-07 14:11:05 -08001894static inline int tcp_inq(struct sock *sk)
1895{
1896 struct tcp_sock *tp = tcp_sk(sk);
1897 int answ;
1898
1899 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1900 answ = 0;
1901 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1902 !tp->urg_data ||
1903 before(tp->urg_seq, tp->copied_seq) ||
1904 !before(tp->urg_seq, tp->rcv_nxt)) {
1905
1906 answ = tp->rcv_nxt - tp->copied_seq;
1907
1908 /* Subtract 1, if FIN was received */
1909 if (answ && sock_flag(sk, SOCK_DONE))
1910 answ--;
1911 } else {
1912 answ = tp->urg_seq - tp->copied_seq;
1913 }
1914
1915 return answ;
1916}
1917
Tom Herbert32035582016-08-28 14:43:18 -07001918int tcp_peek_len(struct socket *sock);
1919
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07001920static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1921{
1922 u16 segs_in;
1923
1924 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1925 tp->segs_in += segs_in;
1926 if (skb->len > tcp_hdrlen(skb))
1927 tp->data_segs_in += segs_in;
1928}
1929
Eric Dumazet9caad862016-04-01 08:52:20 -07001930/*
1931 * TCP listen path runs lockless.
1932 * We forced "struct sock" to be const qualified to make sure
1933 * we don't modify one of its field by mistake.
1934 * Here, we increment sk_drops which is an atomic_t, so we can safely
1935 * make sock writable again.
1936 */
1937static inline void tcp_listendrop(const struct sock *sk)
1938{
1939 atomic_inc(&((struct sock *)sk)->sk_drops);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001940 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
Eric Dumazet9caad862016-04-01 08:52:20 -07001941}
1942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943#endif /* _TCP_H */