blob: e5cf514ba118e688ce3b3da66f696abd47e1d10f [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001/* SPDX-License-Identifier: GPL-2.0-or-later */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the TCP module.
8 *
9 * Version: @(#)tcp.h 1.0.5 05/23/93
10 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070011 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 */
14#ifndef _TCP_H
15#define _TCP_H
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#define FASTRETRANS_DEBUG 1
18
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/list.h>
20#include <linux/tcp.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050021#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/cache.h>
24#include <linux/percpu.h>
Herbert Xufb286bb2005-11-10 13:01:24 -080025#include <linux/skbuff.h>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080026#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000027#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080028#include <linux/ktime.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070029
30#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070031#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070032#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070034#include <net/request_sock.h>
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -070035#include <net/sock_reuseport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <net/sock.h>
37#include <net/snmp.h>
38#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070039#include <net/tcp_states.h>
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -070040#include <net/inet_ecn.h>
Satoru SATOH0c266892009-05-04 11:11:01 -070041#include <net/dst.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/seq_file.h>
Glauber Costa180d8cd2011-12-11 21:47:02 +000044#include <linux/memcontrol.h>
Lawrence Brakmo40304b22017-06-30 20:02:40 -070045#include <linux/bpf-cgroup.h>
Ard Biesheuvel438ac882019-06-19 23:46:28 +020046#include <linux/siphash.h>
Lawrence Brakmo40304b22017-06-30 20:02:40 -070047
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070048extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Eric Dumazetdd24c002008-11-25 21:17:14 -080050extern struct percpu_counter tcp_orphan_count;
Joe Perches5c9f3022013-09-23 11:33:32 -070051void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070054#define MAX_TCP_OPTION_SPACE 40
Eric Dumazet3b4929f2019-05-17 17:17:22 -070055#define TCP_MIN_SND_MSS 48
56#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090058/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090060 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 */
62#define MAX_TCP_WINDOW 32767U
63
64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65#define TCP_MIN_MSS 88U
66
John Heffner5d424d52006-03-20 17:53:41 -080067/* The least MTU to use for probing */
Fan Dudcd8fb82015-03-06 11:18:22 +080068#define TCP_BASE_MSS 1024
John Heffner5d424d52006-03-20 17:53:41 -080069
Fan Du05cbc0d2015-03-06 11:18:24 +080070/* probing interval, default to 10 minutes as per RFC4821 */
71#define TCP_PROBE_INTERVAL 600
72
Fan Du6b58e0a2015-03-06 11:18:23 +080073/* Specify interval when tcp mtu probing will stop */
74#define TCP_PROBE_THRESHOLD 8
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* After receiving this amount of duplicate ACKs fast retransmit starts. */
77#define TCP_FASTRETRANS_THRESH 3
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* Maximal number of ACKs sent quickly to accelerate slow-start. */
80#define TCP_MAX_QUICKACKS 16U
81
Gao Feng589c49c2017-04-04 21:09:48 +080082/* Maximal number of window scale according to RFC1323 */
83#define TCP_MAX_WSCALE 14U
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085/* urg_data states */
86#define TCP_URG_VALID 0x0100
87#define TCP_URG_NOTYET 0x0200
88#define TCP_URG_READ 0x0400
89
90#define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
95 */
96
97#define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
102 */
103
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000104#define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
111 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000113#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
117 * initial RTO.
118 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
126 * TIME-WAIT timer.
127 */
128
129#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
130#if HZ >= 100
131#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132#define TCP_ATO_MIN ((unsigned)(HZ/25))
133#else
134#define TCP_DELACK_MIN 4U
135#define TCP_ATO_MIN 4U
136#endif
137#define TCP_RTO_MAX ((unsigned)(120*HZ))
138#define TCP_RTO_MIN ((unsigned)(HZ/5))
Yuchung Chengbb4d9912017-07-19 15:41:26 -0700139#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000140#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000141#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
146 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
150 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153#define TCP_KEEPALIVE_INTVL (75*HZ)
154
155#define MAX_TCP_KEEPIDLE 32767
156#define MAX_TCP_KEEPINTVL 32767
157#define MAX_TCP_KEEPCNT 127
158#define MAX_TCP_SYNCNT 127
159
160#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
168 */
169#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
172 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173/*
174 * TCP option
175 */
Kenjiro Nakayama105970f2014-10-20 18:15:50 +0900176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177#define TCPOPT_NOP 1 /* Padding */
178#define TCPOPT_EOL 0 /* End of options */
179#define TCPOPT_MSS 2 /* Segment size negotiating */
180#define TCPOPT_WINDOW 3 /* Window scaling */
181#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182#define TCPOPT_SACK 5 /* SACK Block */
183#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800184#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Daniel Lee7f9b8382015-04-06 14:37:26 -0700185#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000186#define TCPOPT_EXP 254 /* Experimental */
187/* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
189 */
190#define TCPOPT_FASTOPEN_MAGIC 0xF989
Ursula Braun60e2a772017-10-25 11:01:45 +0200191#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193/*
194 * TCP option lengths
195 */
196
197#define TCPOLEN_MSS 4
198#define TCPOLEN_WINDOW 3
199#define TCPOLEN_SACK_PERM 2
200#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800201#define TCPOLEN_MD5SIG 18
Daniel Lee7f9b8382015-04-06 14:37:26 -0700202#define TCPOLEN_FASTOPEN_BASE 2
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000203#define TCPOLEN_EXP_FASTOPEN_BASE 4
Ursula Braun60e2a772017-10-25 11:01:45 +0200204#define TCPOLEN_EXP_SMC_BASE 6
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
206/* But this is what stacks really send out. */
207#define TCPOLEN_TSTAMP_ALIGNED 12
208#define TCPOLEN_WSCALE_ALIGNED 4
209#define TCPOLEN_SACKPERM_ALIGNED 4
210#define TCPOLEN_SACK_BASE 2
211#define TCPOLEN_SACK_BASE_ALIGNED 4
212#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800213#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700214#define TCPOLEN_MSS_ALIGNED 4
Ursula Braun60e2a772017-10-25 11:01:45 +0200215#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217/* Flags in tp->nonagle */
218#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800220#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000222/* TCP thin-stream limits */
223#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
224
Jörg Thalheim21603fc2016-01-27 22:54:06 +0100225/* TCP initial congestion window as per rfc6928 */
David S. Miller442b9632011-02-02 17:05:11 -0800226#define TCP_INIT_CWND 10
227
Yuchung Chengcf60af032012-07-19 06:43:09 +0000228/* Bit Flags for sysctl_tcp_fastopen */
229#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000230#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000231#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af032012-07-19 06:43:09 +0000232
Jerry Chu10467162012-08-31 12:29:11 +0000233/* Accept SYN data w/o any cookie option */
234#define TFO_SERVER_COOKIE_NOT_REQD 0x200
235
236/* Force enable TFO on all listeners, i.e., not requiring the
Yuchung Chengcebc5cb2016-08-22 17:17:54 -0700237 * TCP_FASTOPEN socket option.
Jerry Chu10467162012-08-31 12:29:11 +0000238 */
239#define TFO_SERVER_WO_SOCKOPT1 0x400
Jerry Chu10467162012-08-31 12:29:11 +0000240
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243extern int sysctl_tcp_max_orphans;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700244extern long sysctl_tcp_mem[3];
Eric Dumazete20223f2017-10-26 21:54:57 -0700245
Yuchung Chenga0370b32017-01-12 22:11:36 -0800246#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
Priyaranjan Jha1f255692017-11-03 16:38:48 -0700247#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
Yuchung Cheng20b654d2018-05-16 16:40:10 -0700248#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
Yuchung Chenga0370b32017-01-12 22:11:36 -0800249
Eric Dumazet8d987e52010-11-09 23:24:26 +0000250extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800251extern struct percpu_counter tcp_sockets_allocated;
Eric Dumazet06044752017-06-07 13:29:12 -0700252extern unsigned long tcp_memory_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700254/* optimized version of sk_under_memory_pressure() for TCP sockets */
255static inline bool tcp_under_memory_pressure(const struct sock *sk)
256{
Johannes Weinerbaac50b2016-01-14 15:21:17 -0800257 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
258 mem_cgroup_under_socket_pressure(sk->sk_memcg))
Johannes Weinere8056052016-01-14 15:21:14 -0800259 return true;
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700260
261 return tcp_memory_pressure;
262}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
266 */
267
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000268static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800270 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800272#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000275static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
277 return seq3 - seq2 >= seq1 - seq2;
278}
279
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800280static inline bool tcp_out_of_memory(struct sock *sk)
281{
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
284 return true;
285 return false;
286}
287
Eric Dumazeta6c5ea42015-05-15 12:39:26 -0700288void sk_forced_mem_schedule(struct sock *sk, int size);
289
David S. Millerad1af0f2010-08-25 02:27:49 -0700290static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700291{
David S. Millerad1af0f2010-08-25 02:27:49 -0700292 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
293 int orphans = percpu_counter_read_positive(ocp);
294
295 if (orphans << shift > sysctl_tcp_max_orphans) {
296 orphans = percpu_counter_sum_positive(ocp);
297 if (orphans << shift > sysctl_tcp_max_orphans)
298 return true;
299 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700300 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700301}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Joe Perches5c9f3022013-09-23 11:33:32 -0700303bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800304
Florian Westphala0f82f62009-04-19 09:43:48 +0000305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306extern struct proto tcp_prot;
307
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700308#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700309#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700310#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000311#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Joe Perches5c9f3022013-09-23 11:33:32 -0700313void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000314
Stefano Brivio32bbd872018-11-08 12:19:21 +0100315int tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Joe Perches5c9f3022013-09-23 11:33:32 -0700317void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Paolo Abeni74874492017-09-28 15:51:36 +0200319int tcp_v4_early_demux(struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700320int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
Joe Perches5c9f3022013-09-23 11:33:32 -0700322int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800323int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Tom Herbert306b13e2017-07-28 16:22:41 -0700324int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700325int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
326 int flags);
Tom Herbert306b13e2017-07-28 16:22:41 -0700327int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
328 size_t size, int flags);
Dave Watsone3b56162017-06-14 11:37:26 -0700329ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
330 size_t size, int flags);
Joe Perches5c9f3022013-09-23 11:33:32 -0700331void tcp_release_cb(struct sock *sk);
332void tcp_wfree(struct sk_buff *skb);
333void tcp_write_timer_handler(struct sock *sk);
334void tcp_delack_timer_handler(struct sock *sk);
335int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700336int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
Yafang Shao3d97d882018-05-29 23:27:31 +0800337void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700338void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700339int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340void tcp_twsk_destructor(struct sock *sk);
341ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 struct pipe_inode_info *pipe, size_t len,
343 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800344
Yuchung Chenga0496ef22018-07-18 13:56:36 -0700345void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700346static inline void tcp_dec_quickack_mode(struct sock *sk,
347 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700349 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700350
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700351 if (icsk->icsk_ack.quick) {
352 if (pkts >= icsk->icsk_ack.quick) {
353 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700354 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700355 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700356 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700357 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 }
359}
360
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700361#define TCP_ECN_OK 1
362#define TCP_ECN_QUEUE_CWR 2
363#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000364#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700365
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000366enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 TCP_TW_SUCCESS = 0,
368 TCP_TW_RST = 1,
369 TCP_TW_ACK = 2,
370 TCP_TW_SYN = 3
371};
372
373
Joe Perches5c9f3022013-09-23 11:33:32 -0700374enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
375 struct sk_buff *skb,
376 const struct tcphdr *th);
377struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazete0f97592018-02-13 06:14:12 -0800378 struct request_sock *req, bool fastopen,
379 bool *lost_race);
Joe Perches5c9f3022013-09-23 11:33:32 -0700380int tcp_child_process(struct sock *parent, struct sock *child,
381 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400382void tcp_enter_loss(struct sock *sk);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800383void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
Joe Perches5c9f3022013-09-23 11:33:32 -0700384void tcp_clear_retrans(struct tcp_sock *tp);
385void tcp_update_metrics(struct sock *sk);
386void tcp_init_metrics(struct sock *sk);
387void tcp_metrics_init(void);
Soheil Hassas Yeganehd82bae12017-03-15 16:30:45 -0400388bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700389void tcp_close(struct sock *sk, long timeout);
390void tcp_init_sock(struct sock *sk);
Wei Wang27204aa2017-10-04 10:03:44 -0700391void tcp_init_transfer(struct sock *sk, int bpf_op);
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700392__poll_t tcp_poll(struct file *file, struct socket *sock,
393 struct poll_table_struct *wait);
Joe Perches5c9f3022013-09-23 11:33:32 -0700394int tcp_getsockopt(struct sock *sk, int level, int optname,
395 char __user *optval, int __user *optlen);
396int tcp_setsockopt(struct sock *sk, int level, int optname,
397 char __user *optval, unsigned int optlen);
398int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000399 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700400int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000401 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700402void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700403void tcp_syn_ack_timeout(const struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800404int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
405 int flags, int *addr_len);
Eric Dumazetd1361842018-04-16 10:33:35 -0700406int tcp_set_rcvlowat(struct sock *sk, int val);
Eric Dumazet03f45c82018-04-16 10:33:37 -0700407void tcp_data_ready(struct sock *sk);
Yafang Shao340a6f32019-01-17 18:03:14 +0800408#ifdef CONFIG_MMU
Eric Dumazet93ab6cc2018-04-16 10:33:38 -0700409int tcp_mmap(struct file *file, struct socket *sock,
410 struct vm_area_struct *vma);
Yafang Shao340a6f32019-01-17 18:03:14 +0800411#endif
Eric Dumazeteed29f12017-06-07 10:34:36 -0700412void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700413 struct tcp_options_received *opt_rx,
414 int estab, struct tcp_fastopen_cookie *foc);
415const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417/*
418 * TCP v4 functions exported for the inet6 API
419 */
420
Joe Perches5c9f3022013-09-23 11:33:32 -0700421void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400422void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet9cf74902016-02-02 19:31:12 -0800423void tcp_req_err(struct sock *sk, u32 seq, bool abort);
Joe Perches5c9f3022013-09-23 11:33:32 -0700424int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700425struct sock *tcp_create_openreq_child(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -0700426 struct request_sock *req,
427 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100428void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Eric Dumazet0c271712015-09-29 07:42:48 -0700429struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700430 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700431 struct dst_entry *dst,
432 struct request_sock *req_unhash,
433 bool *own_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700434int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
435int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
436int tcp_connect(struct sock *sk);
Eric Dumazetb3d05142016-04-13 22:05:39 -0700437enum tcp_synack_type {
438 TCP_SYNACK_NORMAL,
439 TCP_SYNACK_FASTOPEN,
440 TCP_SYNACK_COOKIE,
441};
Eric Dumazet5d062de2015-09-25 07:39:19 -0700442struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
Joe Perches5c9f3022013-09-23 11:33:32 -0700443 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700444 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -0700445 enum tcp_synack_type synack_type);
Joe Perches5c9f3022013-09-23 11:33:32 -0700446int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000448void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000449int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000450void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452/* From syncookies.c */
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700453struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
454 struct request_sock *req,
Eric Dumazet84b114b2017-05-05 06:56:54 -0700455 struct dst_entry *dst, u32 tsoff);
Joe Perches5c9f3022013-09-23 11:33:32 -0700456int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
457 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700458struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400459#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200460
Eric Dumazet63262312014-03-19 21:02:21 -0700461/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200462 * This counter is used both as a hash input and partially encoded into
463 * the cookie value. A cookie is only validated further if the delta
464 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700465 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200466 * the counter advances immediately after a cookie is generated).
467 */
Eric Dumazet264ea102015-05-14 14:26:56 -0700468#define MAX_SYNCOOKIE_AGE 2
469#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
470#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
471
472/* syncookies: remember time of last synqueue overflow
473 * But do not dirty this field too often (once per second is enough)
Eric Dumazet3f684b42015-09-29 07:42:49 -0700474 * It is racy as we do not hold a lock, but race is very minor.
Eric Dumazet264ea102015-05-14 14:26:56 -0700475 */
Eric Dumazet3f684b42015-09-29 07:42:49 -0700476static inline void tcp_synq_overflow(const struct sock *sk)
Eric Dumazet264ea102015-05-14 14:26:56 -0700477{
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700478 unsigned int last_overflow;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +0200479 unsigned int now = jiffies;
Eric Dumazet264ea102015-05-14 14:26:56 -0700480
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700481 if (sk->sk_reuseport) {
482 struct sock_reuseport *reuse;
483
484 reuse = rcu_dereference(sk->sk_reuseport_cb);
485 if (likely(reuse)) {
486 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
487 if (time_after32(now, last_overflow + HZ))
488 WRITE_ONCE(reuse->synq_overflow_ts, now);
489 return;
490 }
491 }
492
493 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +0200494 if (time_after32(now, last_overflow + HZ))
Eric Dumazet264ea102015-05-14 14:26:56 -0700495 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
496}
497
498/* syncookies: no recent synqueue overflow on this listening socket? */
499static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
500{
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700501 unsigned int last_overflow;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +0200502 unsigned int now = jiffies;
Eric Dumazet264ea102015-05-14 14:26:56 -0700503
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700504 if (sk->sk_reuseport) {
505 struct sock_reuseport *reuse;
506
507 reuse = rcu_dereference(sk->sk_reuseport_cb);
508 if (likely(reuse)) {
509 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
510 return time_after32(now, last_overflow +
511 TCP_SYNCOOKIE_VALID);
512 }
513 }
514
515 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +0200516 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
Eric Dumazet264ea102015-05-14 14:26:56 -0700517}
Florian Westphal8c27bd72013-09-20 22:32:55 +0200518
519static inline u32 tcp_cookie_time(void)
520{
Eric Dumazet63262312014-03-19 21:02:21 -0700521 u64 val = get_jiffies_64();
522
Eric Dumazet264ea102015-05-14 14:26:56 -0700523 do_div(val, TCP_SYNCOOKIE_PERIOD);
Eric Dumazet63262312014-03-19 21:02:21 -0700524 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200525}
526
Joe Perches5c9f3022013-09-23 11:33:32 -0700527u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
528 u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700529__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazet9a568de2017-05-16 14:00:14 -0700530u64 cookie_init_timestamp(struct request_sock *req);
Eric Dumazetf9301032017-06-07 10:34:37 -0700531bool cookie_timestamp_decode(const struct net *net,
532 struct tcp_options_received *opt);
Florian Westphalf1673382014-11-03 17:35:02 +0100533bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100534 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700535
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800536/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700537int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
538 u32 cookie);
539struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100540
Joe Perches5c9f3022013-09-23 11:33:32 -0700541u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
542 const struct tcphdr *th, u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700543__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400544#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545/* tcp_output.c */
546
Joe Perches5c9f3022013-09-23 11:33:32 -0700547void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
548 int nonagle);
Eric Dumazet10d3be52016-04-21 10:55:23 -0700549int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
550int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700551void tcp_retransmit_timer(struct sock *sk);
552void tcp_xmit_retransmit_queue(struct sock *);
553void tcp_simple_retransmit(struct sock *);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800554void tcp_enter_recovery(struct sock *sk, bool ece_ack);
Joe Perches5c9f3022013-09-23 11:33:32 -0700555int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Eric Dumazet75c119a2017-10-05 22:21:27 -0700556enum tcp_queue {
557 TCP_FRAG_IN_WRITE_QUEUE,
558 TCP_FRAG_IN_RTX_QUEUE,
559};
560int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
561 struct sk_buff *skb, u32 len,
562 unsigned int mss_now, gfp_t gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Joe Perches5c9f3022013-09-23 11:33:32 -0700564void tcp_send_probe0(struct sock *);
565void tcp_send_partial(struct sock *);
Eric Dumazete520af42015-05-06 14:26:25 -0700566int tcp_write_wakeup(struct sock *, int mib);
Joe Perches5c9f3022013-09-23 11:33:32 -0700567void tcp_send_fin(struct sock *sk);
568void tcp_send_active_reset(struct sock *sk, gfp_t priority);
569int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700570void tcp_push_one(struct sock *, unsigned int mss_now);
Yuchung Cheng27cde442018-07-18 13:56:35 -0700571void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
Joe Perches5c9f3022013-09-23 11:33:32 -0700572void tcp_send_ack(struct sock *sk);
573void tcp_send_delayed_ack(struct sock *sk);
574void tcp_send_loss_probe(struct sock *sk);
Neal Cardwelled66dfa2017-11-17 21:06:14 -0500575bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
Martin KaFai Laucfea5a62016-04-19 22:39:29 -0700576void tcp_skb_collapse_tstamp(struct sk_buff *skb,
577 const struct sk_buff *next_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
David S. Millera762a982005-07-05 15:18:51 -0700579/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700580void tcp_rearm_rto(struct sock *sk);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700581void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700582void tcp_reset(struct sock *sk);
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -0700583void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
Eric Dumazete3e17b72016-02-06 11:16:28 -0800584void tcp_fin(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700587void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700588static inline void tcp_clear_xmit_timers(struct sock *sk)
589{
Eric Dumazet73a6bab2018-05-10 14:59:43 -0700590 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
Eric Dumazetcf0dd202018-05-17 14:47:24 -0700591 __sock_put(sk);
Eric Dumazet73a6bab2018-05-10 14:59:43 -0700592
Eric Dumazet5d9f4262018-05-17 14:47:26 -0700593 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
594 __sock_put(sk);
595
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700596 inet_csk_clear_xmit_timers(sk);
597}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Joe Perches5c9f3022013-09-23 11:33:32 -0700599unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
600unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000601
602/* Bound MSS / TSO packet size with the half of the window */
603static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
604{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700605 int cutoff;
606
607 /* When peer uses tiny windows, there is no use in packetizing
608 * to sub-MSS pieces for the sake of SWS or making sure there
609 * are enough packets in the pipe for fast recovery.
610 *
611 * On the other hand, for extremely large MSS devices, handling
612 * smaller than MSS windows in this way does make sense.
613 */
Seymour, Shane M2631b792016-06-28 23:06:48 +0000614 if (tp->max_window > TCP_MSS_DEFAULT)
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700615 cutoff = (tp->max_window >> 1);
616 else
617 cutoff = tp->max_window;
618
619 if (cutoff && pktsize > cutoff)
620 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000621 else
622 return pktsize;
623}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300625/* tcp.c */
Eric Dumazet0df48c22015-04-28 15:28:17 -0700626void tcp_get_info(struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628/* Read 'sendfile()'-style from a TCP socket */
Joe Perches5c9f3022013-09-23 11:33:32 -0700629int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
630 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Joe Perches5c9f3022013-09-23 11:33:32 -0700632void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Joe Perches5c9f3022013-09-23 11:33:32 -0700634int tcp_mtu_to_mss(struct sock *sk, int pmtu);
635int tcp_mss_to_mtu(struct sock *sk, int mss);
636void tcp_mtup_init(struct sock *sk);
637void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800638
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000639static inline void tcp_bound_rto(const struct sock *sk)
640{
641 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
642 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
643}
644
645static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
646{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800647 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000648}
649
Florian Westphal31770e32017-08-30 19:24:58 +0200650static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
651{
652 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
653 ntohl(TCP_FLAG_ACK) |
654 snd_wnd);
655}
656
657static inline void tcp_fast_path_on(struct tcp_sock *tp)
658{
659 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
660}
661
662static inline void tcp_fast_path_check(struct sock *sk)
663{
664 struct tcp_sock *tp = tcp_sk(sk);
665
666 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
667 tp->rcv_wnd &&
668 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
669 !tp->urg_data)
670 tcp_fast_path_on(tp);
671}
672
Satoru SATOH0c266892009-05-04 11:11:01 -0700673/* Compute the actual rto_min value */
674static inline u32 tcp_rto_min(struct sock *sk)
675{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400676 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700677 u32 rto_min = TCP_RTO_MIN;
678
679 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
680 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
681 return rto_min;
682}
683
Eric Dumazet740b0f12014-02-26 14:02:48 -0800684static inline u32 tcp_rto_min_us(struct sock *sk)
685{
686 return jiffies_to_usecs(tcp_rto_min(sk));
687}
688
Daniel Borkmann81164412015-01-05 23:57:48 +0100689static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
690{
691 return dst_metric_locked(dst, RTAX_CC_ALGO);
692}
693
Yuchung Chengf6722582015-10-16 21:57:42 -0700694/* Minimum RTT in usec. ~0 means not available. */
695static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
696{
Neal Cardwell64033892016-09-19 23:39:10 -0400697 return minmax_get(&tp->rtt_min);
Yuchung Chengf6722582015-10-16 21:57:42 -0700698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700/* Compute the actual receive window we are currently advertising.
701 * Rcv_nxt can be after the window if our peer push more data
702 * than the offered window.
703 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800704static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
706 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
707
708 if (win < 0)
709 win = 0;
710 return (u32) win;
711}
712
713/* Choose a new window, without checks for shrinking, and without
714 * scaling applied to the result. The caller does these things
715 * if necessary. This is a "raw" window selection.
716 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700717u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Pavel Emelyanovee995282012-04-19 03:40:39 +0000719void tcp_send_window_probe(struct sock *sk);
720
Eric Dumazetec66eda2017-05-16 14:00:01 -0700721/* TCP uses 32bit jiffies to save some space.
722 * Note that this is different from tcp_time_stamp, which
723 * historically has been the same until linux-4.13.
724 */
725#define tcp_jiffies32 ((u32)jiffies)
726
Eric Dumazet9a568de2017-05-16 14:00:14 -0700727/*
728 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
729 * It is no longer tied to jiffies, but to 1 ms clock.
730 * Note: double check if you want to use tcp_jiffies32 instead of this.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700732#define TCP_TS_HZ 1000
733
734static inline u64 tcp_clock_ns(void)
735{
Eric Dumazetfb420d52018-09-28 10:28:44 -0700736 return ktime_get_ns();
Eric Dumazet9a568de2017-05-16 14:00:14 -0700737}
738
739static inline u64 tcp_clock_us(void)
740{
741 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
742}
743
744/* This should only be used in contexts where tp->tcp_mstamp is up to date */
745static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
746{
747 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
748}
749
750/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
751static inline u32 tcp_time_stamp_raw(void)
752{
753 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
754}
755
Eric Dumazet9799ccb2018-09-21 08:51:49 -0700756void tcp_mstamp_refresh(struct tcp_sock *tp);
Eric Dumazet9a568de2017-05-16 14:00:14 -0700757
758static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
759{
760 return max_t(s64, t1 - t0, 0);
761}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700763static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
764{
Eric Dumazetd3edd062018-09-21 08:51:50 -0700765 return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ);
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700766}
767
Eric Dumazet2fd66ff2018-09-21 08:51:47 -0700768/* provide the departure time in us unit */
769static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
770{
Eric Dumazetd3edd062018-09-21 08:51:50 -0700771 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
Eric Dumazet2fd66ff2018-09-21 08:51:47 -0700772}
773
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700774
Changli Gaoa3433f32010-06-12 14:01:43 +0000775#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
776
777#define TCPHDR_FIN 0x01
778#define TCPHDR_SYN 0x02
779#define TCPHDR_RST 0x04
780#define TCPHDR_PSH 0x08
781#define TCPHDR_ACK 0x10
782#define TCPHDR_URG 0x20
783#define TCPHDR_ECE 0x40
784#define TCPHDR_CWR 0x80
785
Daniel Borkmann49213552015-05-19 21:04:22 +0200786#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
787
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800788/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586fa2010-07-15 21:41:00 -0700789 * TCP per-packet control information to the transmission code.
790 * We also store the host-order sequence numbers in here too.
791 * This is 44 bytes if IPV6 is enabled.
792 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 */
794struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 __u32 seq; /* Starting sequence number */
796 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700797 union {
798 /* Note : tcp_tw_isn is used in input path only
799 * (isn chosen by tcp_timewait_state_process())
800 *
Eric Dumazetf69ad292015-06-11 09:15:18 -0700801 * tcp_gso_segs/size are used in write queue only,
802 * cf tcp_skb_pcount()/tcp_skb_mss()
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700803 */
804 __u32 tcp_tw_isn;
Eric Dumazetf69ad292015-06-11 09:15:18 -0700805 struct {
806 u16 tcp_gso_segs;
807 u16 tcp_gso_size;
808 };
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700809 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400810 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000811
Yuchung Cheng713bafe2017-11-08 13:01:26 -0800812 __u8 sacked; /* State flags for SACK. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
814#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
815#define TCPCB_LOST 0x04 /* SKB is lost */
816#define TCPCB_TAGBITS 0x07 /* All tag bits */
Eric Dumazetd3edd062018-09-21 08:51:50 -0700817#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400819#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
820 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000822 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
Soheil Hassas Yeganeh6b084922016-04-02 23:08:08 -0400823 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700824 eor:1, /* Is skb MSG_EOR marked? */
Mike Maloney98aaa912017-08-22 17:08:48 -0400825 has_rxtstamp:1, /* SKB has a RX timestamp */
826 unused:5;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700828 union {
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700829 struct {
Yuchung Chengb9f64822016-09-19 23:39:14 -0400830 /* There is space for up to 24 bytes */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -0400831 __u32 in_flight:30,/* Bytes in flight at transmit */
832 is_app_limited:1, /* cwnd not fully used? */
833 unused:1;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400834 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
835 __u32 delivered;
836 /* start of send pipeline phase */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700837 u64 first_tx_mstamp;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400838 /* when we reached the "delivered" count */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700839 u64 delivered_mstamp;
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700840 } tx; /* only used for outgoing skbs */
841 union {
842 struct inet_skb_parm h4;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700843#if IS_ENABLED(CONFIG_IPV6)
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700844 struct inet6_skb_parm h6;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700845#endif
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700846 } header; /* For incoming skbs */
John Fastabend34f795022017-10-18 07:10:36 -0700847 struct {
John Fastabend34f795022017-10-18 07:10:36 -0700848 __u32 flags;
John Fastabende5cd3ab2018-05-14 10:00:16 -0700849 struct sock *sk_redir;
John Fastabend8108a772017-10-27 09:45:34 -0700850 void *data_end;
John Fastabend34f795022017-10-18 07:10:36 -0700851 } bpf;
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700852 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853};
854
855#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
856
John Fastabend0ea488f2018-07-05 08:50:15 -0700857static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
858{
859 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
860}
Eric Dumazet870c3152014-10-17 09:17:20 -0700861
Daniel Borkmann604326b2018-10-13 02:45:58 +0200862static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
863{
864 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
865}
866
867static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
868{
869 return TCP_SKB_CB(skb)->bpf.sk_redir;
870}
871
872static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
873{
874 TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
875}
876
Eric Dumazet815afe12014-10-18 08:34:37 -0700877#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700878/* This is the variant of inet6_iif() that must be used by TCP,
879 * as TCP moves IP6CB into a different location in skb->cb[]
880 */
881static inline int tcp_v6_iif(const struct sk_buff *skb)
882{
David Ahern24b711e2018-07-19 12:41:18 -0700883 return TCP_SKB_CB(skb)->header.h6.iif;
884}
885
886static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
887{
David Aherna04a4802016-10-16 20:02:52 -0700888 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
David Ahern74b20582016-05-10 11:19:50 -0700889
890 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
Eric Dumazet870c3152014-10-17 09:17:20 -0700891}
David Ahern4297a0e2017-08-07 08:44:21 -0700892
893/* TCP_SKB_CB reference means this can not be used from early demux */
894static inline int tcp_v6_sdif(const struct sk_buff *skb)
895{
896#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
897 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
898 return TCP_SKB_CB(skb)->header.h6.iif;
899#endif
900 return 0;
901}
Eric Dumazet815afe12014-10-18 08:34:37 -0700902#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700903
David Aherna04a4802016-10-16 20:02:52 -0700904static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
905{
906#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
907 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
David Ahernb4d16052017-12-03 09:33:00 -0800908 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
David Aherna04a4802016-10-16 20:02:52 -0700909 return true;
910#endif
911 return false;
912}
913
David Ahern3fa6f612017-08-07 08:44:17 -0700914/* TCP_SKB_CB reference means this can not be used from early demux */
915static inline int tcp_v4_sdif(struct sk_buff *skb)
916{
917#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
918 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
919 return TCP_SKB_CB(skb)->header.h4.iif;
920#endif
921 return 0;
922}
923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924/* Due to TSO, an SKB can be composed of multiple actual
925 * packets. To keep these tracked properly, we use this.
926 */
927static inline int tcp_skb_pcount(const struct sk_buff *skb)
928{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700929 return TCP_SKB_CB(skb)->tcp_gso_segs;
930}
931
932static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
933{
934 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
935}
936
937static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
938{
939 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940}
941
Eric Dumazetf69ad292015-06-11 09:15:18 -0700942/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943static inline int tcp_skb_mss(const struct sk_buff *skb)
944{
Eric Dumazetf69ad292015-06-11 09:15:18 -0700945 return TCP_SKB_CB(skb)->tcp_gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946}
947
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700948static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
949{
950 return likely(!TCP_SKB_CB(skb)->eor);
951}
952
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700953/* Events passed to congestion control interface */
954enum tcp_ca_event {
955 CA_EVENT_TX_START, /* first transmit when no packets in flight */
956 CA_EVENT_CWND_RESTART, /* congestion window restart */
957 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700958 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200959 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
960 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200961};
962
Florian Westphal98900922014-09-26 22:37:35 +0200963/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200964enum tcp_ca_ack_event_flags {
Florian Westphalc1d2b4c2017-08-30 19:24:57 +0200965 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
966 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
967 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700968};
969
970/*
971 * Interface for adding new TCP congestion control handlers
972 */
973#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800974#define TCP_CA_MAX 128
975#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
976
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100977#define TCP_CA_UNSPEC 0
978
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200979/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700980#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200981/* Requires ECN/ECT set on all packets */
982#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700983
Eric Dumazet64f40ff2015-04-28 16:23:48 -0700984union tcp_cc_info;
985
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700986struct ack_sample {
987 u32 pkts_acked;
988 s32 rtt_us;
Lawrence Brakmo6f094b92016-06-08 21:16:44 -0700989 u32 in_flight;
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700990};
991
Yuchung Chengb9f64822016-09-19 23:39:14 -0400992/* A rate sample measures the number of (original/retransmitted) data
993 * packets delivered "delivered" over an interval of time "interval_us".
994 * The tcp_rate.c code fills in the rate sample, and congestion
995 * control modules that define a cong_control function to run at the end
996 * of ACK processing can optionally chose to consult this sample when
997 * setting cwnd and pacing rate.
998 * A sample is invalid if "delivered" or "interval_us" is negative.
999 */
1000struct rate_sample {
Eric Dumazet9a568de2017-05-16 14:00:14 -07001001 u64 prior_mstamp; /* starting timestamp for interval */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001002 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1003 s32 delivered; /* number of packets delivered over interval */
1004 long interval_us; /* time for tp->delivered to incr "delivered" */
Deepti Raghavan4929c942018-07-09 17:53:39 +00001005 u32 snd_interval_us; /* snd interval for delivered packets */
1006 u32 rcv_interval_us; /* rcv interval for delivered packets */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001007 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1008 int losses; /* number of packets marked lost upon ACK */
1009 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1010 u32 prior_in_flight; /* in flight before this ACK */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001011 bool is_app_limited; /* is sample from packet with bubble in pipe? */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001012 bool is_retrans; /* is sample from retransmission? */
Yuchung Chenge4286602018-01-17 12:11:01 -08001013 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001014};
1015
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001016struct tcp_congestion_ops {
1017 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001018 u32 key;
1019 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001020
1021 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001022 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001023 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001024 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001025
1026 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001027 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001028 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -07001029 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001030 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001031 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001032 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001033 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +02001034 /* call when ack arrives (optional) */
1035 void (*in_ack_event)(struct sock *sk, u32 flags);
Anmol Sarma1e0ce2a2017-06-03 17:40:54 +05301036 /* new value of cwnd after loss (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001037 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001038 /* hook for packet ack accounting (optional) */
Lawrence Brakmo756ee172016-05-11 10:02:13 -07001039 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
Eric Dumazetdcb8c9b2018-02-28 14:40:46 -08001040 /* override sysctl_tcp_min_tso_segs */
1041 u32 (*min_tso_segs)(struct sock *sk);
Yuchung Cheng77bfc172016-09-19 23:39:20 -04001042 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1043 u32 (*sndbuf_expand)(struct sock *sk);
Yuchung Chengc0402762016-09-19 23:39:21 -04001044 /* call when packets are delivered to update cwnd and pacing rate,
1045 * after all the ca_state processing. (optional)
1046 */
1047 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -03001048 /* get info for inet_diag (optional) */
Eric Dumazet64f40ff2015-04-28 16:23:48 -07001049 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1050 union tcp_cc_info *info);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001051
1052 char name[TCP_CA_NAME_MAX];
1053 struct module *owner;
1054};
1055
Joe Perches5c9f3022013-09-23 11:33:32 -07001056int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1057void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001058
Florian Westphal55d86942014-09-26 22:37:32 +02001059void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001060void tcp_init_congestion_control(struct sock *sk);
1061void tcp_cleanup_congestion_control(struct sock *sk);
Stephen Hemminger6670e152017-11-14 08:25:49 -08001062int tcp_set_default_congestion_control(struct net *net, const char *name);
1063void tcp_get_default_congestion_control(struct net *net, char *name);
Joe Perches5c9f3022013-09-23 11:33:32 -07001064void tcp_get_available_congestion_control(char *buf, size_t len);
1065void tcp_get_allowed_congestion_control(char *buf, size_t len);
1066int tcp_set_allowed_congestion_control(char *allowed);
Eric Dumazet8d650cd2019-07-18 19:28:14 -07001067int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1068 bool reinit, bool cap_net_admin);
Neal Cardwelle73ebb02015-01-28 20:01:35 -05001069u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1070void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001071
Joe Perches5c9f3022013-09-23 11:33:32 -07001072u32 tcp_reno_ssthresh(struct sock *sk);
Florian Westphale9799182016-11-21 14:18:38 +01001073u32 tcp_reno_undo_cwnd(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -07001074void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -07001075extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001076
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001077struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
Stephen Hemminger6670e152017-11-14 08:25:49 -08001078u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
Daniel Borkmannea697632015-01-05 23:57:47 +01001079#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001080char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +01001081#else
1082static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1083{
1084 return NULL;
1085}
1086#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001087
Daniel Borkmann30e502a2014-09-26 22:37:33 +02001088static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1089{
1090 const struct inet_connection_sock *icsk = inet_csk(sk);
1091
1092 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1093}
1094
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001095static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001096{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001097 struct inet_connection_sock *icsk = inet_csk(sk);
1098
1099 if (icsk->icsk_ca_ops->set_state)
1100 icsk->icsk_ca_ops->set_state(sk, ca_state);
1101 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001102}
1103
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001104static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001105{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001106 const struct inet_connection_sock *icsk = inet_csk(sk);
1107
1108 if (icsk->icsk_ca_ops->cwnd_event)
1109 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001110}
1111
Yuchung Chengb9f64822016-09-19 23:39:14 -04001112/* From tcp_rate.c */
1113void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1114void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1115 struct rate_sample *rs);
1116void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
Yousuk Seungd4761752017-12-07 13:41:34 -08001117 bool is_sack_reneg, struct rate_sample *rs);
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001118void tcp_rate_check_app_limited(struct sock *sk);
Yuchung Chengb9f64822016-09-19 23:39:14 -04001119
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001120/* These functions determine how the current flow behaves in respect of SACK
1121 * handling. SACK is negotiated with the peer, and therefore it can vary
1122 * between different flows.
1123 *
1124 * tcp_is_sack - SACK enabled
1125 * tcp_is_reno - No SACK
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001126 */
1127static inline int tcp_is_sack(const struct tcp_sock *tp)
1128{
Eric Dumazetebeef4b2018-11-27 14:42:00 -08001129 return likely(tp->rx_opt.sack_ok);
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001130}
1131
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001132static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001133{
1134 return !tcp_is_sack(tp);
1135}
1136
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001137static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1138{
1139 return tp->sacked_out + tp->lost_out;
1140}
1141
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142/* This determines how many packets are "in the network" to the best
1143 * of our knowledge. In many cases it is conservative, but where
1144 * detailed information is available from the receiver (via SACK
1145 * blocks etc.) we can make more aggressive calculations.
1146 *
1147 * Use this for decisions involving congestion control, use just
1148 * tp->packets_out to determine if the send queue is empty or not.
1149 *
1150 * Read this equation as:
1151 *
1152 * "Packets sent once on transmission queue" MINUS
1153 * "Packets left network, but not honestly ACKed yet" PLUS
1154 * "Packets fast retransmitted"
1155 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001156static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001158 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159}
1160
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001161#define TCP_INFINITE_SSTHRESH 0x7fffffff
1162
Yuchung Cheng071d5082015-07-09 13:16:29 -07001163static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1164{
Yuchung Cheng76174002015-07-09 13:16:30 -07001165 return tp->snd_cwnd < tp->snd_ssthresh;
Yuchung Cheng071d5082015-07-09 13:16:29 -07001166}
1167
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001168static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1169{
1170 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1171}
1172
Yuchung Cheng684bad12012-09-02 17:38:04 +00001173static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1174{
1175 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1176 (1 << inet_csk(sk)->icsk_ca_state);
1177}
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +00001180 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 * ssthresh.
1182 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001183static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001185 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001186
Yuchung Cheng684bad12012-09-02 17:38:04 +00001187 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 return tp->snd_ssthresh;
1189 else
1190 return max(tp->snd_ssthresh,
1191 ((tp->snd_cwnd >> 1) +
1192 (tp->snd_cwnd >> 2)));
1193}
1194
Ilpo Järvinenb9c45952007-07-27 16:36:17 +03001195/* Use define here intentionally to get WARN_ON location shown at the caller */
1196#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Christoph Paasch5ee2c942014-07-14 16:58:32 +02001198void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001199__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
Neal Cardwell6b5a5c02011-11-21 17:15:14 +00001201/* The maximum number of MSS of available cwnd for which TSO defers
1202 * sending if not using sysctl_tcp_tso_win_divisor.
1203 */
1204static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1205{
1206 return 3;
1207}
1208
Ilpo Järvinen90840defa2007-12-31 04:48:41 -08001209/* Returns end sequence number of the receiver's advertised window */
1210static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1211{
1212 return tp->snd_una + tp->snd_wnd;
1213}
Eric Dumazete114a712014-04-30 11:58:13 -07001214
1215/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1216 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001217 * it was fully used previously. And that's exactly what we do in
1218 * congestion avoidance mode. But in slow start we allow cwnd to grow
1219 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001220 * Example :
1221 * cwnd is 10 (IW10), but application sends 9 frames.
1222 * We allow cwnd to reach 18 when all frames are ACKed.
1223 * This check is safe because it's as aggressive as slow start which already
1224 * risks 100% overshoot. The advantage is that we discourage application to
1225 * either send more filler packets or data to artificially blow up the cwnd
1226 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001227 */
Eric Dumazet24901552014-05-02 21:18:05 -07001228static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001229{
1230 const struct tcp_sock *tp = tcp_sk(sk);
1231
Neal Cardwellca8a2262014-05-22 10:41:08 -04001232 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
Yuchung Cheng071d5082015-07-09 13:16:29 -07001233 if (tcp_in_slow_start(tp))
Neal Cardwellca8a2262014-05-22 10:41:08 -04001234 return tp->snd_cwnd < 2 * tp->max_packets_out;
1235
1236 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001237}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001238
Eric Dumazetcadefe52018-06-20 16:07:35 -04001239/* BBR congestion control needs pacing.
1240 * Same remark for SO_MAX_PACING_RATE.
1241 * sch_fq packet scheduler is efficiently handling pacing,
1242 * but is not always installed/used.
1243 * Return true if TCP stack should pace packets itself.
1244 */
1245static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1246{
1247 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1248}
1249
Eric Dumazet3f80e082018-10-23 11:54:16 -07001250/* Return in jiffies the delay before one skb is sent.
1251 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1252 */
1253static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1254 const struct sk_buff *skb)
1255{
1256 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1257
1258 pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1259
1260 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1261}
1262
1263static inline void tcp_reset_xmit_timer(struct sock *sk,
1264 const int what,
1265 unsigned long when,
1266 const unsigned long max_when,
1267 const struct sk_buff *skb)
1268{
1269 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1270 max_when);
1271}
1272
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001273/* Something is really bad, we could not queue an additional packet,
Eric Dumazet3f80e082018-10-23 11:54:16 -07001274 * because qdisc is full or receiver sent a 0 window, or we are paced.
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001275 * We do not want to add fuel to the fire, or abort too early,
1276 * so make sure the timer we arm now is at least 200ms in the future,
1277 * regardless of current icsk_rto value (as it could be ~2ms)
1278 */
1279static inline unsigned long tcp_probe0_base(const struct sock *sk)
1280{
1281 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1282}
1283
1284/* Variant of inet_csk_rto_backoff() used for zero window probes */
1285static inline unsigned long tcp_probe0_when(const struct sock *sk,
1286 unsigned long max_when)
1287{
1288 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1289
1290 return (unsigned long)min_t(u64, when, max_when);
1291}
1292
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001293static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001295 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
Eric Dumazet3f80e082018-10-23 11:54:16 -07001296 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1297 tcp_probe0_base(sk), TCP_RTO_MAX,
1298 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299}
1300
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001301static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
1303 tp->snd_wl1 = seq;
1304}
1305
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001306static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307{
1308 tp->snd_wl1 = seq;
1309}
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311/*
1312 * Calculate(/check) TCP checksum
1313 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001314static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1315 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
Daniel T. Lee0b13c9b2019-04-21 00:50:42 +09001317 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001320static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
Herbert Xu60476372007-04-09 11:59:39 -07001322 return !skb_csum_unnecessary(skb) &&
Cong Wang6ab6dfa2018-11-12 16:17:00 -08001323 __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324}
1325
Eric Dumazetc9c33212016-08-27 07:37:54 -07001326bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
Eric Dumazetac6e7802016-11-10 13:12:35 -08001327int tcp_filter(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001328void tcp_set_state(struct sock *sk, int state);
Joe Perches5c9f3022013-09-23 11:33:32 -07001329void tcp_done(struct sock *sk);
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09001330int tcp_abort(struct sock *sk, int err);
1331
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001332static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 rx_opt->num_sacks = 0;
1336}
1337
Joe Perches5c9f3022013-09-23 11:33:32 -07001338u32 tcp_default_init_rwnd(u32 mss);
Eric Dumazet6f021c62015-08-21 12:30:00 -07001339void tcp_cwnd_restart(struct sock *sk, s32 delta);
1340
1341static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1342{
Wei Wang1b1fc3f2017-05-05 12:53:23 -07001343 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
Eric Dumazet6f021c62015-08-21 12:30:00 -07001344 struct tcp_sock *tp = tcp_sk(sk);
1345 s32 delta;
1346
Eric Dumazetb510f0d2017-10-26 21:54:59 -07001347 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
Wei Wang1b1fc3f2017-05-05 12:53:23 -07001348 ca_ops->cong_control)
Eric Dumazet6f021c62015-08-21 12:30:00 -07001349 return;
Eric Dumazetd635fbe2017-05-16 14:00:03 -07001350 delta = tcp_jiffies32 - tp->lsndtime;
Eric Dumazet6f021c62015-08-21 12:30:00 -07001351 if (delta > inet_csk(sk)->icsk_rto)
1352 tcp_cwnd_restart(sk, delta);
1353}
Yuchung Cheng85f16522013-06-11 15:35:32 -07001354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355/* Determine a window scaling and initial window to offer. */
Eric Dumazetceef9ab2017-10-27 07:47:24 -07001356void tcp_select_initial_window(const struct sock *sk, int __space,
1357 __u32 mss, __u32 *rcv_wnd,
Joe Perches5c9f3022013-09-23 11:33:32 -07001358 __u32 *window_clamp, int wscale_ok,
1359 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Eric Dumazet94f08932017-10-26 21:55:09 -07001361static inline int tcp_win_from_space(const struct sock *sk, int space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362{
Eric Dumazet94f08932017-10-26 21:55:09 -07001363 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
Gao Fengc4836742017-03-24 07:05:12 +08001364
1365 return tcp_adv_win_scale <= 0 ?
1366 (space>>(-tcp_adv_win_scale)) :
1367 space - (space>>tcp_adv_win_scale);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368}
1369
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001370/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371static inline int tcp_space(const struct sock *sk)
1372{
Eric Dumazet85bdf7d2018-11-27 14:42:02 -08001373 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len -
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001375}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377static inline int tcp_full_space(const struct sock *sk)
1378{
Eric Dumazet94f08932017-10-26 21:55:09 -07001379 return tcp_win_from_space(sk, sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001382extern void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -07001383 const struct sock *sk_listener,
1384 const struct dst_entry *dst);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001385
Joe Perches5c9f3022013-09-23 11:33:32 -07001386void tcp_enter_memory_pressure(struct sock *sk);
Eric Dumazet06044752017-06-07 13:29:12 -07001387void tcp_leave_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1390{
Nikolay Borisovb840d152016-01-07 16:38:45 +02001391 struct net *net = sock_net((struct sock *)tp);
1392
1393 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
1395
1396static inline int keepalive_time_when(const struct tcp_sock *tp)
1397{
Nikolay Borisov13b287e2016-01-07 16:38:43 +02001398 struct net *net = sock_net((struct sock *)tp);
1399
1400 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402
Eric Dumazetdf19a622009-08-28 23:48:54 -07001403static inline int keepalive_probes(const struct tcp_sock *tp)
1404{
Nikolay Borisov9bd68612016-01-07 16:38:44 +02001405 struct net *net = sock_net((struct sock *)tp);
1406
1407 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
Eric Dumazetdf19a622009-08-28 23:48:54 -07001408}
1409
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001410static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1411{
1412 const struct inet_connection_sock *icsk = &tp->inet_conn;
1413
Eric Dumazet70eabf02017-05-16 14:00:07 -07001414 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1415 tcp_jiffies32 - tp->rcv_tstamp);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001416}
1417
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001418static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
Nikolay Borisov1e579ca2016-02-03 09:46:56 +02001420 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001421 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001423 if (fin_timeout < (rto << 2) - (rto >> 1))
1424 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
1426 return fin_timeout;
1427}
1428
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001429static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1430 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001432 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001433 return true;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +02001434 if (unlikely(!time_before32(ktime_get_seconds(),
1435 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001436 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001437 /*
1438 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1439 * then following tcp messages have valid values. Ignore 0 value,
1440 * or else 'negative' tsval might forbid us to accept their packets.
1441 */
1442 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001443 return true;
1444 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001445}
1446
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001447static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1448 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001449{
1450 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001451 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 /* RST segments are not recommended to carry timestamp,
1454 and, if they do, it is recommended to ignore PAWS because
1455 "their cleanup function should take precedence over timestamps."
1456 Certainly, it is mistake. It is necessary to understand the reasons
1457 of this constraint to relax it: if peer reboots, clock may go
1458 out-of-sync and half-open connections will not be reset.
1459 Actually, the problem would be not existing if all
1460 the implementations followed draft about maintaining clock
1461 via reboots. Linux-2.2 DOES NOT!
1462
1463 However, we can relax time bounds for RST segments to MSL.
1464 */
Arnd Bergmanncca9bab2018-07-11 12:16:12 +02001465 if (rst && !time_before32(ktime_get_seconds(),
1466 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001467 return false;
1468 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469}
1470
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001471bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1472 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001473
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001474static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475{
1476 /* See RFC 2012 */
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001477 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1478 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1479 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1480 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481}
1482
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001483/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001484static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001485{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001486 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001487}
1488
1489static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1490{
1491 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001492 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001493}
1494
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001495union tcp_md5_addr {
1496 struct in_addr a4;
1497#if IS_ENABLED(CONFIG_IPV6)
1498 struct in6_addr a6;
1499#endif
1500};
1501
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001502/* - key database */
1503struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001504 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001505 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001506 u8 family; /* AF_INET or AF_INET6 */
1507 union tcp_md5_addr addr;
Ivan Delalande67973182017-06-15 18:07:06 -07001508 u8 prefixlen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001509 u8 key[TCP_MD5SIG_MAXKEYLEN];
1510 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001511};
1512
1513/* - sock block */
1514struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001515 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001516 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001517};
1518
1519/* - pseudo header */
1520struct tcp4_pseudohdr {
1521 __be32 saddr;
1522 __be32 daddr;
1523 __u8 pad;
1524 __u8 protocol;
1525 __be16 len;
1526};
1527
1528struct tcp6_pseudohdr {
1529 struct in6_addr saddr;
1530 struct in6_addr daddr;
1531 __be32 len;
1532 __be32 protocol; /* including padding */
1533};
1534
1535union tcp_md5sum_block {
1536 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001537#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001538 struct tcp6_pseudohdr ip6;
1539#endif
1540};
1541
1542/* - pool: digest algorithm, hash description and scratch buffer */
1543struct tcp_md5sig_pool {
Herbert Xucf80e0e2016-01-24 21:20:23 +08001544 struct ahash_request *md5_req;
Eric Dumazet19689e32016-06-27 18:51:53 +02001545 void *scratch;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001546};
1547
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001548/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001549int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1550 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001551int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
Ivan Delalande67973182017-06-15 18:07:06 -07001552 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1553 gfp_t gfp);
Joe Perches5c9f3022013-09-23 11:33:32 -07001554int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
Ivan Delalande67973182017-06-15 18:07:06 -07001555 int family, u8 prefixlen);
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001556struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001557 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001558
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001559#ifdef CONFIG_TCP_MD5SIG
Eric Dumazet6015c712018-11-27 15:03:21 -08001560#include <linux/jump_label.h>
Eric Dumazet921f9a02019-02-26 09:49:11 -08001561extern struct static_key_false tcp_md5_needed;
Eric Dumazet6015c712018-11-27 15:03:21 -08001562struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
1563 const union tcp_md5_addr *addr,
1564 int family);
1565static inline struct tcp_md5sig_key *
1566tcp_md5_do_lookup(const struct sock *sk,
1567 const union tcp_md5_addr *addr,
1568 int family)
1569{
Eric Dumazet921f9a02019-02-26 09:49:11 -08001570 if (!static_branch_unlikely(&tcp_md5_needed))
Eric Dumazet6015c712018-11-27 15:03:21 -08001571 return NULL;
1572 return __tcp_md5_do_lookup(sk, addr, family);
1573}
1574
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001575#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001576#else
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001577static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001578 const union tcp_md5_addr *addr,
1579 int family)
1580{
1581 return NULL;
1582}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001583#define tcp_twsk_md5_key(twsk) NULL
1584#endif
1585
Joe Perches5c9f3022013-09-23 11:33:32 -07001586bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001587
Joe Perches5c9f3022013-09-23 11:33:32 -07001588struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001589static inline void tcp_put_md5sig_pool(void)
1590{
1591 local_bh_enable();
1592}
Eric Dumazet35790c02010-05-16 00:34:04 -07001593
Joe Perches5c9f3022013-09-23 11:33:32 -07001594int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1595 unsigned int header_len);
1596int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1597 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001598
Jerry Chu10467162012-08-31 12:29:11 +00001599/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001600void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
Yuchung Cheng72685862017-12-12 13:10:40 -08001601 struct tcp_fastopen_cookie *cookie);
Joe Perches5c9f3022013-09-23 11:33:32 -07001602void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -07001603 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1604 u16 try_exp);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001605struct tcp_fastopen_request {
1606 /* Fast Open cookie. Size 0 means a cookie request */
1607 struct tcp_fastopen_cookie cookie;
1608 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001609 size_t size;
1610 int copied; /* queued in tcp_connect() */
Willem de Bruijnf859a442019-01-25 11:17:23 -05001611 struct ubuf_info *uarg;
Yuchung Cheng783237e2012-07-19 06:43:07 +00001612};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001613void tcp_free_fastopen_req(struct tcp_sock *tp);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07001614void tcp_fastopen_destroy_cipher(struct sock *sk);
Haishuang Yan43713842017-09-27 11:35:42 +08001615void tcp_fastopen_ctx_destroy(struct net *net);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07001616int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
Ard Biesheuvel438ac882019-06-19 23:46:28 +02001617 void *primary_key, void *backup_key);
Eric Dumazet61d2bca2016-02-01 21:03:07 -08001618void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
Eric Dumazet7c85af82015-09-24 17:16:05 -07001619struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1620 struct request_sock *req,
Christoph Paasch71c02372017-10-23 13:22:23 -07001621 struct tcp_fastopen_cookie *foc,
1622 const struct dst_entry *dst);
Haishuang Yan43713842017-09-27 11:35:42 +08001623void tcp_fastopen_init_key_once(struct net *net);
Wei Wang065263f2017-01-23 10:59:20 -08001624bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1625 struct tcp_fastopen_cookie *cookie);
Wei Wang19f6d3f2017-01-23 10:59:22 -08001626bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
Ard Biesheuvel438ac882019-06-19 23:46:28 +02001627#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
Jason Baron9092a762019-05-29 12:33:57 -04001628#define TCP_FASTOPEN_KEY_MAX 2
1629#define TCP_FASTOPEN_KEY_BUF_LENGTH \
1630 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
Jerry Chu10467162012-08-31 12:29:11 +00001631
1632/* Fastopen key context */
1633struct tcp_fastopen_context {
Ard Biesheuvel438ac882019-06-19 23:46:28 +02001634 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
Ard Biesheuvelc681eda2019-06-17 10:09:33 +02001635 int num;
1636 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001637};
1638
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001639extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
Wei Wang46c2fa32017-04-20 14:45:47 -07001640void tcp_fastopen_active_disable(struct sock *sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001641bool tcp_fastopen_active_should_disable(struct sock *sk);
1642void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
Yuchung Cheng72685862017-12-12 13:10:40 -08001643void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001644
Jason Baron9092a762019-05-29 12:33:57 -04001645/* Caller needs to wrap with rcu_read_(un)lock() */
1646static inline
1647struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1648{
1649 struct tcp_fastopen_context *ctx;
1650
1651 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1652 if (!ctx)
1653 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1654 return ctx;
1655}
1656
1657static inline
1658bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1659 const struct tcp_fastopen_cookie *orig)
1660{
1661 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1662 orig->len == foc->len &&
1663 !memcmp(orig->val, foc->val, foc->len))
1664 return true;
1665 return false;
1666}
1667
1668static inline
1669int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1670{
Ard Biesheuvelc681eda2019-06-17 10:09:33 +02001671 return ctx->num;
Jason Baron9092a762019-05-29 12:33:57 -04001672}
1673
Francis Yan05b055e2016-11-27 23:07:13 -08001674/* Latencies incurred by various limits for a sender. They are
1675 * chronograph-like stats that are mutually exclusive.
1676 */
1677enum tcp_chrono {
1678 TCP_CHRONO_UNSPEC,
1679 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1680 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1681 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1682 __TCP_CHRONO_MAX,
1683};
1684
1685void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1686void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1687
Eric Dumazete2080072017-10-04 12:59:58 -07001688/* This helper is needed, because skb->tcp_tsorted_anchor uses
1689 * the same memory storage than skb->destructor/_skb_refdst
1690 */
1691static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1692{
1693 skb->destructor = NULL;
1694 skb->_skb_refdst = 0UL;
1695}
1696
1697#define tcp_skb_tsorted_save(skb) { \
1698 unsigned long _save = skb->_skb_refdst; \
1699 skb->_skb_refdst = 0UL;
1700
1701#define tcp_skb_tsorted_restore(skb) \
1702 skb->_skb_refdst = _save; \
1703}
1704
Eric Dumazetac3f09b2017-10-05 22:21:22 -07001705void tcp_write_queue_purge(struct sock *sk);
David S. Millerfe067e82007-03-07 12:12:44 -08001706
Eric Dumazet75c119a2017-10-05 22:21:27 -07001707static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1708{
1709 return skb_rb_first(&sk->tcp_rtx_queue);
1710}
1711
Eric Dumazetb6171582019-07-19 11:52:33 -07001712static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1713{
1714 return skb_rb_last(&sk->tcp_rtx_queue);
1715}
1716
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001717static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001718{
David S. Millercd07a8e2008-09-23 00:50:13 -07001719 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001720}
1721
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001722static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001723{
David S. Millercd07a8e2008-09-23 00:50:13 -07001724 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001725}
1726
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001727#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001728 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001729
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001730static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001731{
Eric Dumazet75c119a2017-10-05 22:21:27 -07001732 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001733}
1734
David S. Millercd07a8e2008-09-23 00:50:13 -07001735static inline bool tcp_skb_is_last(const struct sock *sk,
1736 const struct sk_buff *skb)
1737{
1738 return skb_queue_is_last(&sk->sk_write_queue, skb);
1739}
1740
Eric Dumazet75c119a2017-10-05 22:21:27 -07001741static inline bool tcp_write_queue_empty(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001742{
Eric Dumazet75c119a2017-10-05 22:21:27 -07001743 return skb_queue_empty(&sk->sk_write_queue);
1744}
1745
1746static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1747{
1748 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1749}
1750
1751static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1752{
1753 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08001754}
1755
David S. Millerfe067e82007-03-07 12:12:44 -08001756static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1757{
Eric Dumazeta43e0522019-02-26 09:49:10 -08001758 __skb_queue_tail(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001759
1760 /* Queue it, remembering where we must start sending. */
Eric Dumazet50895b92017-11-14 21:02:19 -08001761 if (sk->sk_write_queue.next == skb)
Francis Yan0f872302016-11-27 23:07:14 -08001762 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
David S. Millerfe067e82007-03-07 12:12:44 -08001763}
1764
David S. Miller43f59c82008-09-21 21:28:51 -07001765/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001766static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1767 struct sk_buff *skb,
1768 struct sock *sk)
1769{
David S. Miller43f59c82008-09-21 21:28:51 -07001770 __skb_queue_before(&sk->sk_write_queue, skb, new);
David S. Millerfe067e82007-03-07 12:12:44 -08001771}
1772
1773static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1774{
Eric Dumazet4a269812017-10-11 13:27:29 -07001775 tcp_skb_tsorted_anchor_cleanup(skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001776 __skb_unlink(skb, &sk->sk_write_queue);
1777}
1778
Eric Dumazet75c119a2017-10-05 22:21:27 -07001779void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1780
1781static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001782{
Eric Dumazet75c119a2017-10-05 22:21:27 -07001783 tcp_skb_tsorted_anchor_cleanup(skb);
1784 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1785}
1786
1787static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1788{
1789 list_del(&skb->tcp_tsorted_anchor);
1790 tcp_rtx_queue_unlink(skb, sk);
1791 sk_wmem_free_skb(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001792}
1793
Krishna Kumar12d50c42009-12-08 22:26:13 +00001794static inline void tcp_push_pending_frames(struct sock *sk)
1795{
1796 if (tcp_send_head(sk)) {
1797 struct tcp_sock *tp = tcp_sk(sk);
1798
1799 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1800 }
1801}
1802
Neal Cardwellecb97192012-02-27 17:52:52 -05001803/* Start sequence of the skb just after the highest skb with SACKed
1804 * bit, valid only if sacked_out > 0 or when the caller has ensured
1805 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001806 */
1807static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1808{
1809 if (!tp->sacked_out)
1810 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001811
1812 if (tp->highest_sack == NULL)
1813 return tp->snd_nxt;
1814
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001815 return TCP_SKB_CB(tp->highest_sack)->seq;
1816}
1817
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001818static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1819{
Eric Dumazet50895b92017-11-14 21:02:19 -08001820 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001821}
1822
1823static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1824{
1825 return tcp_sk(sk)->highest_sack;
1826}
1827
1828static inline void tcp_highest_sack_reset(struct sock *sk)
1829{
Eric Dumazet50895b92017-11-14 21:02:19 -08001830 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001831}
1832
Eric Dumazet2b7cda92017-10-30 23:08:20 -07001833/* Called when old skb is about to be deleted and replaced by new skb */
1834static inline void tcp_highest_sack_replace(struct sock *sk,
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001835 struct sk_buff *old,
1836 struct sk_buff *new)
1837{
Eric Dumazet2b7cda92017-10-30 23:08:20 -07001838 if (old == tcp_highest_sack(sk))
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001839 tcp_sk(sk)->highest_sack = new;
1840}
1841
Florian Westphalb1f0a0e2015-12-21 21:29:24 +01001842/* This helper checks if socket has IP_TRANSPARENT set */
1843static inline bool inet_sk_transparent(const struct sock *sk)
1844{
1845 switch (sk->sk_state) {
1846 case TCP_TIME_WAIT:
1847 return inet_twsk(sk)->tw_transparent;
1848 case TCP_NEW_SYN_RECV:
1849 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1850 }
1851 return inet_sk(sk)->transparent;
1852}
1853
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001854/* Determines whether this is a thin stream (which may suffer from
1855 * increased latency). Used to trigger latency-reducing mechanisms.
1856 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001857static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001858{
1859 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1860}
1861
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862/* /proc */
1863enum tcp_seq_states {
1864 TCP_SEQ_STATE_LISTENING,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866};
1867
Christoph Hellwig37d849b2018-04-11 09:31:28 +02001868void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1869void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1870void tcp_seq_stop(struct seq_file *seq, void *v);
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001873 sa_family_t family;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874};
1875
1876struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001877 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 enum tcp_seq_states state;
1879 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001880 int bucket, offset, sbucket, num;
Tom Herberta8b690f2010-06-07 00:43:42 -07001881 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882};
1883
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001884extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001885extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001886
Joe Perches5c9f3022013-09-23 11:33:32 -07001887void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001888
Eric Dumazet28be6e02013-10-18 10:36:17 -07001889struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001890 netdev_features_t features);
David Millerd4546c22018-06-24 14:13:49 +09001891struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001892int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001893
Joe Perches5c9f3022013-09-23 11:33:32 -07001894void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001895
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001896static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1897{
Nikolay Borisov4979f2d2016-02-03 09:46:57 +02001898 struct net *net = sock_net((struct sock *)tp);
1899 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001900}
1901
Eric Dumazeta74f0fa2018-12-04 07:58:17 -08001902/* @wake is one when sk_stream_write_space() calls us.
1903 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1904 * This mimics the strategy used in sock_def_write_space().
1905 */
1906static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001907{
1908 const struct tcp_sock *tp = tcp_sk(sk);
1909 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1910
Eric Dumazeta74f0fa2018-12-04 07:58:17 -08001911 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001912}
1913
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001914#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001915int tcp4_proc_init(void);
1916void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001917#endif
1918
Eric Dumazetea3bea32015-09-25 07:39:23 -07001919int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001920int tcp_conn_request(struct request_sock_ops *rsk_ops,
1921 const struct tcp_request_sock_ops *af_ops,
1922 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001923
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001924/* TCP af-specific functions */
1925struct tcp_sock_af_ops {
1926#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001927 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001928 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001929 int (*calc_md5_hash)(char *location,
1930 const struct tcp_md5sig_key *md5,
1931 const struct sock *sk,
1932 const struct sk_buff *skb);
1933 int (*md5_parse)(struct sock *sk,
Ivan Delalande8917a772017-06-15 18:07:07 -07001934 int optname,
Eric Dumazet39f8e582015-03-24 15:58:55 -07001935 char __user *optval,
1936 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001937#endif
1938};
1939
1940struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001941 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001942#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001943 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001944 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07001945 int (*calc_md5_hash) (char *location,
1946 const struct tcp_md5sig_key *md5,
1947 const struct sock *sk,
1948 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001949#endif
Eric Dumazetb40cf182015-09-25 07:39:08 -07001950 void (*init_req)(struct request_sock *req,
1951 const struct sock *sk_listener,
Octavian Purdila16bea702014-06-25 17:09:53 +03001952 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001953#ifdef CONFIG_SYN_COOKIES
Eric Dumazet3f684b42015-09-29 07:42:49 -07001954 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001955 __u16 *mss);
1956#endif
Eric Dumazetf9646292015-09-29 07:42:50 -07001957 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
Soheil Hassas Yeganeh4396e462017-03-15 16:30:46 -04001958 const struct request_sock *req);
Eric Dumazet84b114b2017-05-05 06:56:54 -07001959 u32 (*init_seq)(const struct sk_buff *skb);
Eric Dumazet5d2ed052017-06-07 10:34:39 -07001960 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
Eric Dumazet0f935db2015-09-25 07:39:21 -07001961 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001962 struct flowi *fl, struct request_sock *req,
Eric Dumazetdc6ef6b2015-10-16 13:00:01 -07001963 struct tcp_fastopen_cookie *foc,
Eric Dumazetb3d05142016-04-13 22:05:39 -07001964 enum tcp_synack_type synack_type);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001965};
1966
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001967#ifdef CONFIG_SYN_COOKIES
1968static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001969 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001970 __u16 *mss)
1971{
Eric Dumazet3f684b42015-09-29 07:42:49 -07001972 tcp_synq_overflow(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07001973 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
Eric Dumazet3f684b42015-09-29 07:42:49 -07001974 return ops->cookie_init_seq(skb, mss);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001975}
1976#else
1977static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07001978 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001979 __u16 *mss)
1980{
1981 return 0;
1982}
1983#endif
1984
Joe Perches5c9f3022013-09-23 11:33:32 -07001985int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001986
Joe Perches5c9f3022013-09-23 11:33:32 -07001987void tcp_v4_init(void);
1988void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001989
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07001990/* tcp_recovery.c */
Yuchung Chengd716bfd2018-05-16 16:40:13 -07001991void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
Yuchung Cheng6ac06ec2018-05-16 16:40:12 -07001992void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
Yuchung Chengb8fef652018-05-16 16:40:16 -07001993extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1994 u32 reo_wnd);
Eric Dumazet128eda82017-04-25 10:15:34 -07001995extern void tcp_rack_mark_lost(struct sock *sk);
Yuchung Cheng1d0833d2017-01-12 22:11:34 -08001996extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
Eric Dumazet9a568de2017-05-16 14:00:14 -07001997 u64 xmit_time);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -08001998extern void tcp_rack_reo_timeout(struct sock *sk);
Priyaranjan Jha1f255692017-11-03 16:38:48 -07001999extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07002000
Neal Cardwelle1a10ef2017-08-03 09:19:52 -04002001/* At how many usecs into the future should the RTO fire? */
2002static inline s64 tcp_rto_delta_us(const struct sock *sk)
2003{
Eric Dumazet75c119a2017-10-05 22:21:27 -07002004 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
Neal Cardwelle1a10ef2017-08-03 09:19:52 -04002005 u32 rto = inet_csk(sk)->icsk_rto;
Eric Dumazet2fd66ff2018-09-21 08:51:47 -07002006 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
Neal Cardwelle1a10ef2017-08-03 09:19:52 -04002007
2008 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2009}
2010
Cong Wange25f8662014-10-15 14:33:21 -07002011/*
2012 * Save and compile IPv4 options, return a pointer to it
2013 */
Paolo Abeni91ed1e62017-08-03 18:07:06 +02002014static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2015 struct sk_buff *skb)
Cong Wange25f8662014-10-15 14:33:21 -07002016{
2017 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2018 struct ip_options_rcu *dopt = NULL;
2019
Cong Wang461b74c2014-10-15 14:33:22 -07002020 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07002021 int opt_size = sizeof(*dopt) + opt->optlen;
2022
2023 dopt = kmalloc(opt_size, GFP_ATOMIC);
Paolo Abeni91ed1e62017-08-03 18:07:06 +02002024 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
Cong Wange25f8662014-10-15 14:33:21 -07002025 kfree(dopt);
2026 dopt = NULL;
2027 }
2028 }
2029 return dopt;
2030}
2031
Eric Dumazet98781962015-02-03 18:31:53 -08002032/* locally generated TCP pure ACKs have skb->truesize == 2
2033 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2034 * This is much faster than dissecting the packet to find out.
2035 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2036 */
2037static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2038{
2039 return skb->truesize == 2;
2040}
2041
2042static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2043{
2044 skb->truesize = 2;
2045}
2046
Tom Herbert473bd232016-03-07 14:11:05 -08002047static inline int tcp_inq(struct sock *sk)
2048{
2049 struct tcp_sock *tp = tcp_sk(sk);
2050 int answ;
2051
2052 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2053 answ = 0;
2054 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2055 !tp->urg_data ||
2056 before(tp->urg_seq, tp->copied_seq) ||
2057 !before(tp->urg_seq, tp->rcv_nxt)) {
2058
2059 answ = tp->rcv_nxt - tp->copied_seq;
2060
2061 /* Subtract 1, if FIN was received */
2062 if (answ && sock_flag(sk, SOCK_DONE))
2063 answ--;
2064 } else {
2065 answ = tp->urg_seq - tp->copied_seq;
2066 }
2067
2068 return answ;
2069}
2070
Tom Herbert32035582016-08-28 14:43:18 -07002071int tcp_peek_len(struct socket *sock);
2072
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07002073static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2074{
2075 u16 segs_in;
2076
2077 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2078 tp->segs_in += segs_in;
2079 if (skb->len > tcp_hdrlen(skb))
2080 tp->data_segs_in += segs_in;
2081}
2082
Eric Dumazet9caad862016-04-01 08:52:20 -07002083/*
2084 * TCP listen path runs lockless.
2085 * We forced "struct sock" to be const qualified to make sure
2086 * we don't modify one of its field by mistake.
2087 * Here, we increment sk_drops which is an atomic_t, so we can safely
2088 * make sock writable again.
2089 */
2090static inline void tcp_listendrop(const struct sock *sk)
2091{
2092 atomic_inc(&((struct sock *)sk)->sk_drops);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07002093 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
Eric Dumazet9caad862016-04-01 08:52:20 -07002094}
2095
Eric Dumazet218af592017-05-16 04:24:36 -07002096enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2097
Dave Watson734942c2017-06-14 11:37:14 -07002098/*
2099 * Interface for adding Upper Level Protocols over TCP
2100 */
2101
2102#define TCP_ULP_NAME_MAX 16
2103#define TCP_ULP_MAX 128
2104#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2105
2106struct tcp_ulp_ops {
2107 struct list_head list;
2108
2109 /* initialize ulp */
2110 int (*init)(struct sock *sk);
2111 /* cleanup ulp */
2112 void (*release)(struct sock *sk);
2113
2114 char name[TCP_ULP_NAME_MAX];
2115 struct module *owner;
2116};
2117int tcp_register_ulp(struct tcp_ulp_ops *type);
2118void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2119int tcp_set_ulp(struct sock *sk, const char *name);
2120void tcp_get_available_ulp(char *buf, size_t len);
2121void tcp_cleanup_ulp(struct sock *sk);
2122
Daniel Borkmann037b0b82018-08-16 21:49:06 +02002123#define MODULE_ALIAS_TCP_ULP(name) \
2124 __MODULE_INFO(alias, alias_userspace, name); \
2125 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2126
Daniel Borkmann604326b2018-10-13 02:45:58 +02002127struct sk_msg;
2128struct sk_psock;
2129
2130int tcp_bpf_init(struct sock *sk);
2131void tcp_bpf_reinit(struct sock *sk);
2132int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2133 int flags);
2134int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2135 int nonblock, int flags, int *addr_len);
2136int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
John Fastabend02c558b2018-10-16 11:08:04 -07002137 struct msghdr *msg, int len, int flags);
Daniel Borkmann604326b2018-10-13 02:45:58 +02002138
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002139/* Call BPF_SOCK_OPS program that returns an int. If the return value
2140 * is < 0, then the BPF op failed (for example if the loaded BPF
2141 * program does not support the chosen operation or there is no BPF
2142 * program loaded).
2143 */
2144#ifdef CONFIG_BPF
Lawrence Brakmode525be2018-01-25 16:14:09 -08002145static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002146{
2147 struct bpf_sock_ops_kern sock_ops;
2148 int ret;
2149
Lawrence Brakmob73042b2018-01-25 16:14:08 -08002150 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
Lawrence Brakmof19397a2017-12-01 10:15:04 -08002151 if (sk_fullsock(sk)) {
2152 sock_ops.is_fullsock = 1;
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002153 sock_owned_by_me(sk);
Lawrence Brakmof19397a2017-12-01 10:15:04 -08002154 }
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002155
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002156 sock_ops.sk = sk;
2157 sock_ops.op = op;
Lawrence Brakmode525be2018-01-25 16:14:09 -08002158 if (nargs > 0)
2159 memcpy(sock_ops.args, args, nargs * sizeof(*args));
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002160
2161 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2162 if (ret == 0)
2163 ret = sock_ops.reply;
2164 else
2165 ret = -1;
2166 return ret;
2167}
Lawrence Brakmode525be2018-01-25 16:14:09 -08002168
2169static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2170{
2171 u32 args[2] = {arg1, arg2};
2172
2173 return tcp_call_bpf(sk, op, 2, args);
2174}
2175
2176static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2177 u32 arg3)
2178{
2179 u32 args[3] = {arg1, arg2, arg3};
2180
2181 return tcp_call_bpf(sk, op, 3, args);
2182}
2183
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002184#else
Lawrence Brakmode525be2018-01-25 16:14:09 -08002185static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002186{
2187 return -EPERM;
2188}
Lawrence Brakmode525be2018-01-25 16:14:09 -08002189
2190static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2191{
2192 return -EPERM;
2193}
2194
2195static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2196 u32 arg3)
2197{
2198 return -EPERM;
2199}
2200
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002201#endif
2202
Lawrence Brakmo8550f322017-06-30 20:02:42 -07002203static inline u32 tcp_timeout_init(struct sock *sk)
2204{
2205 int timeout;
2206
Lawrence Brakmode525be2018-01-25 16:14:09 -08002207 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
Lawrence Brakmo8550f322017-06-30 20:02:42 -07002208
2209 if (timeout <= 0)
2210 timeout = TCP_TIMEOUT_INIT;
2211 return timeout;
2212}
2213
Lawrence Brakmo13d3b1e2017-06-30 20:02:44 -07002214static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2215{
2216 int rwnd;
2217
Lawrence Brakmode525be2018-01-25 16:14:09 -08002218 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
Lawrence Brakmo13d3b1e2017-06-30 20:02:44 -07002219
2220 if (rwnd < 0)
2221 rwnd = 0;
2222 return rwnd;
2223}
Lawrence Brakmo91b5b212017-06-30 20:02:49 -07002224
2225static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2226{
Lawrence Brakmode525be2018-01-25 16:14:09 -08002227 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
Lawrence Brakmo91b5b212017-06-30 20:02:49 -07002228}
Ursula Braun60e2a772017-10-25 11:01:45 +02002229
Stanislav Fomichev23729ff2019-07-02 09:13:56 -07002230static inline void tcp_bpf_rtt(struct sock *sk)
2231{
Arnd Bergmannbef8e262019-07-08 14:57:21 +02002232 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
Stanislav Fomichev23729ff2019-07-02 09:13:56 -07002233 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2234}
2235
Ursula Braun60e2a772017-10-25 11:01:45 +02002236#if IS_ENABLED(CONFIG_SMC)
2237extern struct static_key_false tcp_have_smc;
2238#endif
Ilya Lesokhin6dac1522018-04-30 10:16:10 +03002239
2240#if IS_ENABLED(CONFIG_TLS_DEVICE)
2241void clean_acked_data_enable(struct inet_connection_sock *icsk,
2242 void (*cad)(struct sock *sk, u32 ack_seq));
2243void clean_acked_data_disable(struct inet_connection_sock *icsk);
Jakub Kicinski494bc1d2019-05-08 16:46:14 -07002244void clean_acked_data_flush(void);
Ilya Lesokhin6dac1522018-04-30 10:16:10 +03002245#endif
2246
Eric Dumazeta842fe12019-06-12 11:57:25 -07002247DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2248static inline void tcp_add_tx_delay(struct sk_buff *skb,
2249 const struct tcp_sock *tp)
2250{
2251 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2252 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2253}
2254
Eric Dumazetd6fb3962019-06-13 21:22:35 -07002255/* Compute Earliest Departure Time for some control packets
2256 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2257 */
2258static inline u64 tcp_transmit_time(const struct sock *sk)
Eric Dumazeta842fe12019-06-12 11:57:25 -07002259{
2260 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2261 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2262 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2263
Eric Dumazetd6fb3962019-06-13 21:22:35 -07002264 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
Eric Dumazeta842fe12019-06-12 11:57:25 -07002265 }
Eric Dumazetd6fb3962019-06-13 21:22:35 -07002266 return 0;
Eric Dumazeta842fe12019-06-12 11:57:25 -07002267}
2268
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269#endif /* _TCP_H */