blob: db9f828e9d1ee4546951a408935b6c5f90851a93 [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>
William Allen Simpson435cf552009-12-02 18:17:05 +000026#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080027#include <linux/ktime.h>
Eric Dumazet05e22e82020-06-19 12:12:34 -070028#include <linux/indirect_call_wrapper.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>
Mat Martineau85712482020-01-09 07:59:20 -080042#include <net/mptcp.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/seq_file.h>
Glauber Costa180d8cd2011-12-11 21:47:02 +000045#include <linux/memcontrol.h>
Lawrence Brakmo40304b22017-06-30 20:02:40 -070046#include <linux/bpf-cgroup.h>
Ard Biesheuvel438ac882019-06-19 23:46:28 +020047#include <linux/siphash.h>
Lawrence Brakmo40304b22017-06-30 20:02:40 -070048
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070049extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Eric Dumazet19757ce2021-10-14 06:41:26 -070051DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
52int tcp_orphan_count_sum(void);
53
Joe Perches5c9f3022013-09-23 11:33:32 -070054void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Eric Dumazet9bacd252020-04-17 07:10:23 -070056#define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070057#define MAX_TCP_OPTION_SPACE 40
Eric Dumazet3b4929f2019-05-17 17:17:22 -070058#define TCP_MIN_SND_MSS 48
59#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090061/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090063 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 */
65#define MAX_TCP_WINDOW 32767U
66
67/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68#define TCP_MIN_MSS 88U
69
Josh Hunt1555e6f2019-08-07 19:52:30 -040070/* The initial MTU to use for probing */
Fan Dudcd8fb82015-03-06 11:18:22 +080071#define TCP_BASE_MSS 1024
John Heffner5d424d52006-03-20 17:53:41 -080072
Fan Du05cbc0d2015-03-06 11:18:24 +080073/* probing interval, default to 10 minutes as per RFC4821 */
74#define TCP_PROBE_INTERVAL 600
75
Fan Du6b58e0a2015-03-06 11:18:23 +080076/* Specify interval when tcp mtu probing will stop */
77#define TCP_PROBE_THRESHOLD 8
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* After receiving this amount of duplicate ACKs fast retransmit starts. */
80#define TCP_FASTRETRANS_THRESH 3
81
Linus Torvalds1da177e2005-04-16 15:20:36 -070082/* Maximal number of ACKs sent quickly to accelerate slow-start. */
83#define TCP_MAX_QUICKACKS 16U
84
Gao Feng589c49c2017-04-04 21:09:48 +080085/* Maximal number of window scale according to RFC1323 */
86#define TCP_MAX_WSCALE 14U
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/* urg_data states */
89#define TCP_URG_VALID 0x0100
90#define TCP_URG_NOTYET 0x0200
91#define TCP_URG_READ 0x0400
92
93#define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
98 */
99
100#define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
105 */
106
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000107#define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
114 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000116#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
120 * initial RTO.
121 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
129 * TIME-WAIT timer.
130 */
Cambda Zhuf0628c52020-04-24 16:06:16 +0800131#define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
134#if HZ >= 100
135#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
136#define TCP_ATO_MIN ((unsigned)(HZ/25))
137#else
138#define TCP_DELACK_MIN 4U
139#define TCP_ATO_MIN 4U
140#endif
141#define TCP_RTO_MAX ((unsigned)(120*HZ))
142#define TCP_RTO_MIN ((unsigned)(HZ/5))
Yuchung Chengbb4d9912017-07-19 15:41:26 -0700143#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000144#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000145#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
146 * used as a fallback RTO for the
147 * initial data transmission if no
148 * valid RTT sample has been acquired,
149 * most likely due to retrans in 3WHS.
150 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
152#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
153 * for local resources.
154 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
156#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
157#define TCP_KEEPALIVE_INTVL (75*HZ)
158
159#define MAX_TCP_KEEPIDLE 32767
160#define MAX_TCP_KEEPINTVL 32767
161#define MAX_TCP_KEEPCNT 127
162#define MAX_TCP_SYNCNT 127
163
164#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
167#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
168 * after this time. It should be equal
169 * (or greater than) TCP_TIMEWAIT_LEN
170 * to provide reliability equal to one
171 * provided by timewait state.
172 */
173#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
174 * timestamps. It must be less than
175 * minimal timewait lifetime.
176 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177/*
178 * TCP option
179 */
Kenjiro Nakayama105970f2014-10-20 18:15:50 +0900180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181#define TCPOPT_NOP 1 /* Padding */
182#define TCPOPT_EOL 0 /* End of options */
183#define TCPOPT_MSS 2 /* Segment size negotiating */
184#define TCPOPT_WINDOW 3 /* Window scaling */
185#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
186#define TCPOPT_SACK 5 /* SACK Block */
187#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800188#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Mat Martineauc74a39c2020-01-09 07:59:17 -0800189#define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
Daniel Lee7f9b8382015-04-06 14:37:26 -0700190#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000191#define TCPOPT_EXP 254 /* Experimental */
192/* Magic number to be after the option value for sharing TCP
193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
194 */
195#define TCPOPT_FASTOPEN_MAGIC 0xF989
Ursula Braun60e2a772017-10-25 11:01:45 +0200196#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198/*
199 * TCP option lengths
200 */
201
202#define TCPOLEN_MSS 4
203#define TCPOLEN_WINDOW 3
204#define TCPOLEN_SACK_PERM 2
205#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800206#define TCPOLEN_MD5SIG 18
Daniel Lee7f9b8382015-04-06 14:37:26 -0700207#define TCPOLEN_FASTOPEN_BASE 2
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000208#define TCPOLEN_EXP_FASTOPEN_BASE 4
Ursula Braun60e2a772017-10-25 11:01:45 +0200209#define TCPOLEN_EXP_SMC_BASE 6
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211/* But this is what stacks really send out. */
212#define TCPOLEN_TSTAMP_ALIGNED 12
213#define TCPOLEN_WSCALE_ALIGNED 4
214#define TCPOLEN_SACKPERM_ALIGNED 4
215#define TCPOLEN_SACK_BASE 2
216#define TCPOLEN_SACK_BASE_ALIGNED 4
217#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800218#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700219#define TCPOLEN_MSS_ALIGNED 4
Ursula Braun60e2a772017-10-25 11:01:45 +0200220#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222/* Flags in tp->nonagle */
223#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
224#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800225#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000227/* TCP thin-stream limits */
228#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
229
Jörg Thalheim21603fc2016-01-27 22:54:06 +0100230/* TCP initial congestion window as per rfc6928 */
David S. Miller442b9632011-02-02 17:05:11 -0800231#define TCP_INIT_CWND 10
232
Yuchung Chengcf60af032012-07-19 06:43:09 +0000233/* Bit Flags for sysctl_tcp_fastopen */
234#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000235#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000236#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af032012-07-19 06:43:09 +0000237
Jerry Chu10467162012-08-31 12:29:11 +0000238/* Accept SYN data w/o any cookie option */
239#define TFO_SERVER_COOKIE_NOT_REQD 0x200
240
241/* Force enable TFO on all listeners, i.e., not requiring the
Yuchung Chengcebc5cb2016-08-22 17:17:54 -0700242 * TCP_FASTOPEN socket option.
Jerry Chu10467162012-08-31 12:29:11 +0000243 */
244#define TFO_SERVER_WO_SOCKOPT1 0x400
Jerry Chu10467162012-08-31 12:29:11 +0000245
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248extern int sysctl_tcp_max_orphans;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700249extern long sysctl_tcp_mem[3];
Eric Dumazete20223f2017-10-26 21:54:57 -0700250
Yuchung Chenga0370b32017-01-12 22:11:36 -0800251#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
Priyaranjan Jha1f255692017-11-03 16:38:48 -0700252#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
Yuchung Cheng20b654d2018-05-16 16:40:10 -0700253#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
Yuchung Chenga0370b32017-01-12 22:11:36 -0800254
Eric Dumazet8d987e52010-11-09 23:24:26 +0000255extern atomic_long_t tcp_memory_allocated;
Eric Dumazet0defbb02022-06-08 23:34:08 -0700256DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
257
Eric Dumazet17483762008-11-25 21:16:35 -0800258extern struct percpu_counter tcp_sockets_allocated;
Eric Dumazet06044752017-06-07 13:29:12 -0700259extern unsigned long tcp_memory_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700261/* optimized version of sk_under_memory_pressure() for TCP sockets */
262static inline bool tcp_under_memory_pressure(const struct sock *sk)
263{
Johannes Weinerbaac50b2016-01-14 15:21:17 -0800264 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
265 mem_cgroup_under_socket_pressure(sk->sk_memcg))
Johannes Weinere8056052016-01-14 15:21:14 -0800266 return true;
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700267
Eric Dumazet1f142c12019-10-09 15:10:15 -0700268 return READ_ONCE(tcp_memory_pressure);
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700269}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 * The next routines deal with comparing 32 bit unsigned ints
272 * and worry about wraparound (automatic with unsigned arithmetic).
273 */
274
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000275static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800277 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800279#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
281/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000282static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283{
284 return seq3 - seq2 >= seq1 - seq2;
285}
286
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800287static inline bool tcp_out_of_memory(struct sock *sk)
288{
289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
291 return true;
292 return false;
293}
294
Talal Ahmad03271f32021-10-29 22:05:41 -0400295static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
296{
297 sk_wmem_queued_add(sk, -skb->truesize);
Talal Ahmad9b65b172021-11-02 22:58:44 -0400298 if (!skb_zcopy_pure(skb))
299 sk_mem_uncharge(sk, skb->truesize);
300 else
301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
Talal Ahmad03271f32021-10-29 22:05:41 -0400302 __kfree_skb(skb);
303}
304
Eric Dumazeta6c5ea42015-05-15 12:39:26 -0700305void sk_forced_mem_schedule(struct sock *sk, int size);
306
Joe Perches5c9f3022013-09-23 11:33:32 -0700307bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800308
Florian Westphala0f82f62009-04-19 09:43:48 +0000309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310extern struct proto tcp_prot;
311
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700312#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Eric Dumazet13415e42016-04-27 16:44:43 -0700313#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700314#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000315#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Joe Perches5c9f3022013-09-23 11:33:32 -0700317void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000318
Stefano Brivio32bbd872018-11-08 12:19:21 +0100319int tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Joe Perches5c9f3022013-09-23 11:33:32 -0700321void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Paolo Abeni74874492017-09-28 15:51:36 +0200323int tcp_v4_early_demux(struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700324int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Eric Dumazet27728ba2021-10-27 13:19:18 -0700326void tcp_remove_empty_skb(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700327int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800328int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Tom Herbert306b13e2017-07-28 16:22:41 -0700329int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
Benjamin Hesmans3242abe2022-09-26 16:27:37 -0700330int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
331 size_t size, struct ubuf_info *uarg);
Joe Perches5c9f3022013-09-23 11:33:32 -0700332int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
333 int flags);
Tom Herbert306b13e2017-07-28 16:22:41 -0700334int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
335 size_t size, int flags);
Dave Watsone3b56162017-06-14 11:37:26 -0700336ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
337 size_t size, int flags);
Mat Martineau35b2c322020-01-09 07:59:21 -0800338int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
339void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
340 int size_goal);
Joe Perches5c9f3022013-09-23 11:33:32 -0700341void tcp_release_cb(struct sock *sk);
342void tcp_wfree(struct sk_buff *skb);
343void tcp_write_timer_handler(struct sock *sk);
344void tcp_delack_timer_handler(struct sock *sk);
345int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
Eric Dumazet72ab4a82015-09-29 07:42:41 -0700346int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
Yafang Shao3d97d882018-05-29 23:27:31 +0800347void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700348void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700349int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
350void tcp_twsk_destructor(struct sock *sk);
Kuniyuki Iwashimaedc12f02022-09-07 18:10:21 -0700351void tcp_twsk_purge(struct list_head *net_exit_list, int family);
Joe Perches5c9f3022013-09-23 11:33:32 -0700352ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
353 struct pipe_inode_info *pipe, size_t len,
354 unsigned int flags);
Eric Dumazetf8dd3b82021-10-25 15:13:40 -0700355struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
356 bool force_schedule);
Jens Axboe9c55e012007-11-06 23:30:13 -0800357
Yuchung Chenga0496ef22018-07-18 13:56:36 -0700358void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700359static inline void tcp_dec_quickack_mode(struct sock *sk,
360 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700362 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700363
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700364 if (icsk->icsk_ack.quick) {
365 if (pkts >= icsk->icsk_ack.quick) {
366 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700367 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700368 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700369 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700370 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
372}
373
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700374#define TCP_ECN_OK 1
375#define TCP_ECN_QUEUE_CWR 2
376#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000377#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700378
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000379enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 TCP_TW_SUCCESS = 0,
381 TCP_TW_RST = 1,
382 TCP_TW_ACK = 2,
383 TCP_TW_SYN = 3
384};
385
386
Joe Perches5c9f3022013-09-23 11:33:32 -0700387enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
388 struct sk_buff *skb,
389 const struct tcphdr *th);
390struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
Eric Dumazete0f97592018-02-13 06:14:12 -0800391 struct request_sock *req, bool fastopen,
392 bool *lost_race);
Joe Perches5c9f3022013-09-23 11:33:32 -0700393int tcp_child_process(struct sock *parent, struct sock *child,
394 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400395void tcp_enter_loss(struct sock *sk);
Yuchung Cheng7e901ee2020-10-30 18:34:12 -0700396void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
Joe Perches5c9f3022013-09-23 11:33:32 -0700397void tcp_clear_retrans(struct tcp_sock *tp);
398void tcp_update_metrics(struct sock *sk);
399void tcp_init_metrics(struct sock *sk);
400void tcp_metrics_init(void);
Soheil Hassas Yeganehd82bae12017-03-15 16:30:45 -0400401bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
Paolo Abeni77c3c952020-11-16 10:48:04 +0100402void __tcp_close(struct sock *sk, long timeout);
Joe Perches5c9f3022013-09-23 11:33:32 -0700403void tcp_close(struct sock *sk, long timeout);
404void tcp_init_sock(struct sock *sk);
Martin KaFai Lau72be0fe2020-08-20 12:00:39 -0700405void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
Linus Torvaldsa11e1d42018-06-28 09:43:44 -0700406__poll_t tcp_poll(struct file *file, struct socket *sock,
407 struct poll_table_struct *wait);
Martin KaFai Lau273b7f02022-09-01 17:29:18 -0700408int do_tcp_getsockopt(struct sock *sk, int level,
409 int optname, sockptr_t optval, sockptr_t optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700410int tcp_getsockopt(struct sock *sk, int level, int optname,
411 char __user *optval, int __user *optlen);
Stanislav Fomichev9cacf812021-01-15 08:34:59 -0800412bool tcp_bpf_bypass_getsockopt(int level, int optname);
Martin KaFai Lau0c751f72022-08-16 23:18:19 -0700413int do_tcp_setsockopt(struct sock *sk, int level, int optname,
414 sockptr_t optval, unsigned int optlen);
Christoph Hellwiga7b75c52020-07-23 08:09:07 +0200415int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
416 unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700417void tcp_set_keepalive(struct sock *sk, int val);
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700418void tcp_syn_ack_timeout(const struct request_sock *req);
Oliver Hartkoppec095262022-04-11 14:49:55 +0200419int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
Ying Xue1b784142015-03-02 15:37:48 +0800420 int flags, int *addr_len);
Eric Dumazetd1361842018-04-16 10:33:35 -0700421int tcp_set_rcvlowat(struct sock *sk, int val);
Prankur guptacb811102020-12-02 13:31:51 -0800422int tcp_set_window_clamp(struct sock *sk, int val);
Florian Westphal892bfd32021-06-03 16:24:31 -0700423void tcp_update_recv_tstamps(struct sk_buff *skb,
424 struct scm_timestamping_internal *tss);
425void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
426 struct scm_timestamping_internal *tss);
Eric Dumazet03f45c82018-04-16 10:33:37 -0700427void tcp_data_ready(struct sock *sk);
Yafang Shao340a6f32019-01-17 18:03:14 +0800428#ifdef CONFIG_MMU
Eric Dumazet93ab6cc2018-04-16 10:33:38 -0700429int tcp_mmap(struct file *file, struct socket *sock,
430 struct vm_area_struct *vma);
Yafang Shao340a6f32019-01-17 18:03:14 +0800431#endif
Eric Dumazeteed29f12017-06-07 10:34:36 -0700432void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700433 struct tcp_options_received *opt_rx,
434 int estab, struct tcp_fastopen_cookie *foc);
435const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437/*
Petar Penkov9349d602019-07-29 09:59:14 -0700438 * BPF SKB-less helpers
439 */
440u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
441 struct tcphdr *th, u32 *cookie);
442u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
443 struct tcphdr *th, u32 *cookie);
Maxim Mikityanskiy33bf9882022-06-15 16:48:44 +0300444u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
Petar Penkov9349d602019-07-29 09:59:14 -0700445u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
446 const struct tcp_request_sock_ops *af_ops,
447 struct sock *sk, struct tcphdr *th);
448/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 * TCP v4 functions exported for the inet6 API
450 */
451
Joe Perches5c9f3022013-09-23 11:33:32 -0700452void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400453void tcp_v4_mtu_reduced(struct sock *sk);
Eric Dumazet9cf74902016-02-02 19:31:12 -0800454void tcp_req_err(struct sock *sk, u32 seq, bool abort);
Eric Dumazetd2924562020-05-27 17:34:58 -0700455void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
Joe Perches5c9f3022013-09-23 11:33:32 -0700456int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
Eric Dumazetc28c6f02015-09-29 07:42:47 -0700457struct sock *tcp_create_openreq_child(const struct sock *sk,
Joe Perches5c9f3022013-09-23 11:33:32 -0700458 struct request_sock *req,
459 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100460void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Eric Dumazet0c271712015-09-29 07:42:48 -0700461struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -0700462 struct request_sock *req,
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700463 struct dst_entry *dst,
464 struct request_sock *req_unhash,
465 bool *own_req);
Joe Perches5c9f3022013-09-23 11:33:32 -0700466int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
467int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
468int tcp_connect(struct sock *sk);
Eric Dumazetb3d05142016-04-13 22:05:39 -0700469enum tcp_synack_type {
470 TCP_SYNACK_NORMAL,
471 TCP_SYNACK_FASTOPEN,
472 TCP_SYNACK_COOKIE,
473};
Eric Dumazet5d062de2015-09-25 07:39:19 -0700474struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
Joe Perches5c9f3022013-09-23 11:33:32 -0700475 struct request_sock *req,
Eric Dumazetca6fb062015-10-02 11:43:35 -0700476 struct tcp_fastopen_cookie *foc,
Martin KaFai Lau331fca42020-08-20 12:00:52 -0700477 enum tcp_synack_type synack_type,
478 struct sk_buff *syn_skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700479int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000481void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000482int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000483void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485/* From syncookies.c */
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700486struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
487 struct request_sock *req,
Eric Dumazet84b114b2017-05-05 06:56:54 -0700488 struct dst_entry *dst, u32 tsoff);
Joe Perches5c9f3022013-09-23 11:33:32 -0700489int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
490 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700491struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Florian Westphal6fc8c822020-07-30 21:25:55 +0200492struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
Eric Dumazetba5a4fd2022-04-24 13:35:09 -0700493 const struct tcp_request_sock_ops *af_ops,
Florian Westphal6fc8c822020-07-30 21:25:55 +0200494 struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400495#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200496
Eric Dumazet63262312014-03-19 21:02:21 -0700497/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200498 * This counter is used both as a hash input and partially encoded into
499 * the cookie value. A cookie is only validated further if the delta
500 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700501 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200502 * the counter advances immediately after a cookie is generated).
503 */
Eric Dumazet264ea102015-05-14 14:26:56 -0700504#define MAX_SYNCOOKIE_AGE 2
505#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
506#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
507
508/* syncookies: remember time of last synqueue overflow
509 * But do not dirty this field too often (once per second is enough)
Eric Dumazet3f684b42015-09-29 07:42:49 -0700510 * It is racy as we do not hold a lock, but race is very minor.
Eric Dumazet264ea102015-05-14 14:26:56 -0700511 */
Eric Dumazet3f684b42015-09-29 07:42:49 -0700512static inline void tcp_synq_overflow(const struct sock *sk)
Eric Dumazet264ea102015-05-14 14:26:56 -0700513{
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700514 unsigned int last_overflow;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +0200515 unsigned int now = jiffies;
Eric Dumazet264ea102015-05-14 14:26:56 -0700516
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700517 if (sk->sk_reuseport) {
518 struct sock_reuseport *reuse;
519
520 reuse = rcu_dereference(sk->sk_reuseport_cb);
521 if (likely(reuse)) {
522 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
Guillaume Nault04d26e72019-12-06 12:38:36 +0100523 if (!time_between32(now, last_overflow,
524 last_overflow + HZ))
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700525 WRITE_ONCE(reuse->synq_overflow_ts, now);
526 return;
527 }
528 }
529
Guillaume Nault721c8da2019-12-06 12:38:49 +0100530 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
Guillaume Nault04d26e72019-12-06 12:38:36 +0100531 if (!time_between32(now, last_overflow, last_overflow + HZ))
Guillaume Nault721c8da2019-12-06 12:38:49 +0100532 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
Eric Dumazet264ea102015-05-14 14:26:56 -0700533}
534
535/* syncookies: no recent synqueue overflow on this listening socket? */
536static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
537{
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700538 unsigned int last_overflow;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +0200539 unsigned int now = jiffies;
Eric Dumazet264ea102015-05-14 14:26:56 -0700540
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700541 if (sk->sk_reuseport) {
542 struct sock_reuseport *reuse;
543
544 reuse = rcu_dereference(sk->sk_reuseport_cb);
545 if (likely(reuse)) {
546 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
Guillaume Naultcb44a082019-12-06 12:38:43 +0100547 return !time_between32(now, last_overflow - HZ,
548 last_overflow +
549 TCP_SYNCOOKIE_VALID);
Martin KaFai Lau40a1227e2018-08-08 01:01:21 -0700550 }
551 }
552
Guillaume Nault721c8da2019-12-06 12:38:49 +0100553 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
Guillaume Naultcb44a082019-12-06 12:38:43 +0100554
555 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
556 * then we're under synflood. However, we have to use
557 * 'last_overflow - HZ' as lower bound. That's because a concurrent
558 * tcp_synq_overflow() could update .ts_recent_stamp after we read
559 * jiffies but before we store .ts_recent_stamp into last_overflow,
560 * which could lead to rejecting a valid syncookie.
561 */
562 return !time_between32(now, last_overflow - HZ,
563 last_overflow + TCP_SYNCOOKIE_VALID);
Eric Dumazet264ea102015-05-14 14:26:56 -0700564}
Florian Westphal8c27bd72013-09-20 22:32:55 +0200565
566static inline u32 tcp_cookie_time(void)
567{
Eric Dumazet63262312014-03-19 21:02:21 -0700568 u64 val = get_jiffies_64();
569
Eric Dumazet264ea102015-05-14 14:26:56 -0700570 do_div(val, TCP_SYNCOOKIE_PERIOD);
Eric Dumazet63262312014-03-19 21:02:21 -0700571 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200572}
573
Joe Perches5c9f3022013-09-23 11:33:32 -0700574u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
575 u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700576__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazet200ecef2019-11-07 11:51:18 -0800577u64 cookie_init_timestamp(struct request_sock *req, u64 now);
Eric Dumazetf9301032017-06-07 10:34:37 -0700578bool cookie_timestamp_decode(const struct net *net,
579 struct tcp_options_received *opt);
Florian Westphalf1673382014-11-03 17:35:02 +0100580bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100581 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700582
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800583/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700584int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
585 u32 cookie);
586struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100587
Joe Perches5c9f3022013-09-23 11:33:32 -0700588u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
589 const struct tcphdr *th, u16 *mssp);
Eric Dumazet3f684b42015-09-29 07:42:49 -0700590__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400591#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/* tcp_output.c */
593
Paolo Abeni04d88252021-09-22 19:26:40 +0200594void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
595void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -0700596void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
597 int nonagle);
Eric Dumazet10d3be52016-04-21 10:55:23 -0700598int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
599int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
Joe Perches5c9f3022013-09-23 11:33:32 -0700600void tcp_retransmit_timer(struct sock *sk);
601void tcp_xmit_retransmit_queue(struct sock *);
602void tcp_simple_retransmit(struct sock *);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -0800603void tcp_enter_recovery(struct sock *sk, bool ece_ack);
Joe Perches5c9f3022013-09-23 11:33:32 -0700604int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Eric Dumazet75c119a2017-10-05 22:21:27 -0700605enum tcp_queue {
606 TCP_FRAG_IN_WRITE_QUEUE,
607 TCP_FRAG_IN_RTX_QUEUE,
608};
609int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
610 struct sk_buff *skb, u32 len,
611 unsigned int mss_now, gfp_t gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Joe Perches5c9f3022013-09-23 11:33:32 -0700613void tcp_send_probe0(struct sock *);
614void tcp_send_partial(struct sock *);
Eric Dumazete520af42015-05-06 14:26:25 -0700615int tcp_write_wakeup(struct sock *, int mib);
Joe Perches5c9f3022013-09-23 11:33:32 -0700616void tcp_send_fin(struct sock *sk);
617void tcp_send_active_reset(struct sock *sk, gfp_t priority);
618int tcp_send_synack(struct sock *);
Joe Perches5c9f3022013-09-23 11:33:32 -0700619void tcp_push_one(struct sock *, unsigned int mss_now);
Yuchung Cheng27cde442018-07-18 13:56:35 -0700620void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
Joe Perches5c9f3022013-09-23 11:33:32 -0700621void tcp_send_ack(struct sock *sk);
622void tcp_send_delayed_ack(struct sock *sk);
623void tcp_send_loss_probe(struct sock *sk);
Neal Cardwelled66dfa2017-11-17 21:06:14 -0500624bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
Martin KaFai Laucfea5a62016-04-19 22:39:29 -0700625void tcp_skb_collapse_tstamp(struct sk_buff *skb,
626 const struct sk_buff *next_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
David S. Millera762a982005-07-05 15:18:51 -0700628/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700629void tcp_rearm_rto(struct sock *sk);
Yuchung Cheng0f1c28a2015-09-18 11:36:14 -0700630void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
Florian Westphal049fe382020-12-10 14:25:03 -0800631void tcp_reset(struct sock *sk, struct sk_buff *skb);
Yuchung Cheng4f41b1c2015-10-16 21:57:47 -0700632void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
Eric Dumazete3e17b72016-02-06 11:16:28 -0800633void tcp_fin(struct sock *sk);
Eric Dumazet4bfe7442022-04-24 17:34:07 -0700634void tcp_check_space(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700637void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700638static inline void tcp_clear_xmit_timers(struct sock *sk)
639{
Eric Dumazet73a6bab2018-05-10 14:59:43 -0700640 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
Eric Dumazetcf0dd202018-05-17 14:47:24 -0700641 __sock_put(sk);
Eric Dumazet73a6bab2018-05-10 14:59:43 -0700642
Eric Dumazet5d9f4262018-05-17 14:47:26 -0700643 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
644 __sock_put(sk);
645
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700646 inet_csk_clear_xmit_timers(sk);
647}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Joe Perches5c9f3022013-09-23 11:33:32 -0700649unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
650unsigned int tcp_current_mss(struct sock *sk);
Enke Chen344db932021-01-22 11:13:06 -0800651u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000652
653/* Bound MSS / TSO packet size with the half of the window */
654static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
655{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700656 int cutoff;
657
658 /* When peer uses tiny windows, there is no use in packetizing
659 * to sub-MSS pieces for the sake of SWS or making sure there
660 * are enough packets in the pipe for fast recovery.
661 *
662 * On the other hand, for extremely large MSS devices, handling
663 * smaller than MSS windows in this way does make sense.
664 */
Seymour, Shane M2631b792016-06-28 23:06:48 +0000665 if (tp->max_window > TCP_MSS_DEFAULT)
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700666 cutoff = (tp->max_window >> 1);
667 else
668 cutoff = tp->max_window;
669
670 if (cutoff && pktsize > cutoff)
671 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000672 else
673 return pktsize;
674}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300676/* tcp.c */
Eric Dumazet0df48c22015-04-28 15:28:17 -0700677void tcp_get_info(struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679/* Read 'sendfile()'-style from a TCP socket */
Joe Perches5c9f3022013-09-23 11:33:32 -0700680int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
681 sk_read_actor_t recv_actor);
Cong Wang965b57b2022-06-15 09:20:12 -0700682int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
Jakub Kicinski3f92a642022-07-22 16:50:31 -0700683struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
684void tcp_read_done(struct sock *sk, size_t len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Joe Perches5c9f3022013-09-23 11:33:32 -0700686void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
Joe Perches5c9f3022013-09-23 11:33:32 -0700688int tcp_mtu_to_mss(struct sock *sk, int pmtu);
689int tcp_mss_to_mtu(struct sock *sk, int mss);
690void tcp_mtup_init(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800691
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000692static inline void tcp_bound_rto(const struct sock *sk)
693{
694 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
695 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
696}
697
698static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
699{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800700 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000701}
702
Florian Westphal31770e32017-08-30 19:24:58 +0200703static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
704{
Paolo Abeni71158bb2021-06-30 13:42:13 +0200705 /* mptcp hooks are only on the slow path */
706 if (sk_is_mptcp((struct sock *)tp))
707 return;
708
Florian Westphal31770e32017-08-30 19:24:58 +0200709 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
710 ntohl(TCP_FLAG_ACK) |
711 snd_wnd);
712}
713
714static inline void tcp_fast_path_on(struct tcp_sock *tp)
715{
716 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
717}
718
719static inline void tcp_fast_path_check(struct sock *sk)
720{
721 struct tcp_sock *tp = tcp_sk(sk);
722
723 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
724 tp->rcv_wnd &&
725 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
726 !tp->urg_data)
727 tcp_fast_path_on(tp);
728}
729
Satoru SATOH0c266892009-05-04 11:11:01 -0700730/* Compute the actual rto_min value */
731static inline u32 tcp_rto_min(struct sock *sk)
732{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400733 const struct dst_entry *dst = __sk_dst_get(sk);
Martin KaFai Lauca584ba2020-08-20 12:00:27 -0700734 u32 rto_min = inet_csk(sk)->icsk_rto_min;
Satoru SATOH0c266892009-05-04 11:11:01 -0700735
736 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
737 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
738 return rto_min;
739}
740
Eric Dumazet740b0f12014-02-26 14:02:48 -0800741static inline u32 tcp_rto_min_us(struct sock *sk)
742{
743 return jiffies_to_usecs(tcp_rto_min(sk));
744}
745
Daniel Borkmann81164412015-01-05 23:57:48 +0100746static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
747{
748 return dst_metric_locked(dst, RTAX_CC_ALGO);
749}
750
Yuchung Chengf6722582015-10-16 21:57:42 -0700751/* Minimum RTT in usec. ~0 means not available. */
752static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
753{
Neal Cardwell64033892016-09-19 23:39:10 -0400754 return minmax_get(&tp->rtt_min);
Yuchung Chengf6722582015-10-16 21:57:42 -0700755}
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757/* Compute the actual receive window we are currently advertising.
758 * Rcv_nxt can be after the window if our peer push more data
759 * than the offered window.
760 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800761static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762{
763 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
764
765 if (win < 0)
766 win = 0;
767 return (u32) win;
768}
769
770/* Choose a new window, without checks for shrinking, and without
771 * scaling applied to the result. The caller does these things
772 * if necessary. This is a "raw" window selection.
773 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700774u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Pavel Emelyanovee995282012-04-19 03:40:39 +0000776void tcp_send_window_probe(struct sock *sk);
777
Eric Dumazetec66eda2017-05-16 14:00:01 -0700778/* TCP uses 32bit jiffies to save some space.
779 * Note that this is different from tcp_time_stamp, which
780 * historically has been the same until linux-4.13.
781 */
782#define tcp_jiffies32 ((u32)jiffies)
783
Eric Dumazet9a568de2017-05-16 14:00:14 -0700784/*
785 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
786 * It is no longer tied to jiffies, but to 1 ms clock.
787 * Note: double check if you want to use tcp_jiffies32 instead of this.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700789#define TCP_TS_HZ 1000
790
791static inline u64 tcp_clock_ns(void)
792{
Eric Dumazetfb420d52018-09-28 10:28:44 -0700793 return ktime_get_ns();
Eric Dumazet9a568de2017-05-16 14:00:14 -0700794}
795
796static inline u64 tcp_clock_us(void)
797{
798 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
799}
800
801/* This should only be used in contexts where tp->tcp_mstamp is up to date */
802static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
803{
804 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
805}
806
Eric Dumazet200ecef2019-11-07 11:51:18 -0800807/* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
808static inline u32 tcp_ns_to_ts(u64 ns)
809{
810 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
811}
812
Eric Dumazet9a568de2017-05-16 14:00:14 -0700813/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
814static inline u32 tcp_time_stamp_raw(void)
815{
Eric Dumazet200ecef2019-11-07 11:51:18 -0800816 return tcp_ns_to_ts(tcp_clock_ns());
Eric Dumazet9a568de2017-05-16 14:00:14 -0700817}
818
Eric Dumazet9799ccb2018-09-21 08:51:49 -0700819void tcp_mstamp_refresh(struct tcp_sock *tp);
Eric Dumazet9a568de2017-05-16 14:00:14 -0700820
821static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
822{
823 return max_t(s64, t1 - t0, 0);
824}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700826static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
827{
Eric Dumazet200ecef2019-11-07 11:51:18 -0800828 return tcp_ns_to_ts(skb->skb_mstamp_ns);
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700829}
830
Eric Dumazet2fd66ff2018-09-21 08:51:47 -0700831/* provide the departure time in us unit */
832static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
833{
Eric Dumazetd3edd062018-09-21 08:51:50 -0700834 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
Eric Dumazet2fd66ff2018-09-21 08:51:47 -0700835}
836
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700837
Changli Gaoa3433f32010-06-12 14:01:43 +0000838#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
839
840#define TCPHDR_FIN 0x01
841#define TCPHDR_SYN 0x02
842#define TCPHDR_RST 0x04
843#define TCPHDR_PSH 0x08
844#define TCPHDR_ACK 0x10
845#define TCPHDR_URG 0x20
846#define TCPHDR_ECE 0x40
847#define TCPHDR_CWR 0x80
848
Daniel Borkmann49213552015-05-19 21:04:22 +0200849#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
850
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800851/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586fa2010-07-15 21:41:00 -0700852 * TCP per-packet control information to the transmission code.
853 * We also store the host-order sequence numbers in here too.
854 * This is 44 bytes if IPV6 is enabled.
855 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 */
857struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 __u32 seq; /* Starting sequence number */
859 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700860 union {
861 /* Note : tcp_tw_isn is used in input path only
862 * (isn chosen by tcp_timewait_state_process())
863 *
Eric Dumazetf69ad292015-06-11 09:15:18 -0700864 * tcp_gso_segs/size are used in write queue only,
865 * cf tcp_skb_pcount()/tcp_skb_mss()
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700866 */
867 __u32 tcp_tw_isn;
Eric Dumazetf69ad292015-06-11 09:15:18 -0700868 struct {
869 u16 tcp_gso_segs;
870 u16 tcp_gso_size;
871 };
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700872 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400873 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000874
Yuchung Cheng713bafe2017-11-08 13:01:26 -0800875 __u8 sacked; /* State flags for SACK. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
877#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
878#define TCPCB_LOST 0x04 /* SKB is lost */
879#define TCPCB_TAGBITS 0x07 /* All tag bits */
Eric Dumazetd3edd062018-09-21 08:51:50 -0700880#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400882#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
883 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000885 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
Soheil Hassas Yeganeh6b084922016-04-02 23:08:08 -0400886 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700887 eor:1, /* Is skb MSG_EOR marked? */
Mike Maloney98aaa912017-08-22 17:08:48 -0400888 has_rxtstamp:1, /* SKB has a RX timestamp */
889 unused:5;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700891 union {
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700892 struct {
Yuchung Cheng40bc6062021-09-23 21:17:07 +0000893#define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
Yuchung Chengb9f64822016-09-19 23:39:14 -0400894 /* There is space for up to 24 bytes */
Yuchung Cheng40bc6062021-09-23 21:17:07 +0000895 __u32 is_app_limited:1, /* cwnd not fully used? */
896 delivered_ce:20,
897 unused:11;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400898 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
899 __u32 delivered;
900 /* start of send pipeline phase */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700901 u64 first_tx_mstamp;
Yuchung Chengb9f64822016-09-19 23:39:14 -0400902 /* when we reached the "delivered" count */
Eric Dumazet9a568de2017-05-16 14:00:14 -0700903 u64 delivered_mstamp;
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700904 } tx; /* only used for outgoing skbs */
905 union {
906 struct inet_skb_parm h4;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700907#if IS_ENABLED(CONFIG_IPV6)
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700908 struct inet6_skb_parm h6;
Eric Dumazet971f10e2014-09-27 09:50:57 -0700909#endif
Lawrence Brakmob75803d2016-05-06 20:35:35 -0700910 } header; /* For incoming skbs */
911 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912};
913
914#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
915
Eric Dumazet9b9e2f22020-06-23 15:31:13 -0700916extern const struct inet_connection_sock_af_ops ipv4_specific;
917
Eric Dumazet815afe12014-10-18 08:34:37 -0700918#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700919/* This is the variant of inet6_iif() that must be used by TCP,
920 * as TCP moves IP6CB into a different location in skb->cb[]
921 */
922static inline int tcp_v6_iif(const struct sk_buff *skb)
923{
David Ahern24b711e2018-07-19 12:41:18 -0700924 return TCP_SKB_CB(skb)->header.h6.iif;
925}
926
927static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
928{
David Aherna04a4802016-10-16 20:02:52 -0700929 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
David Ahern74b20582016-05-10 11:19:50 -0700930
931 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
Eric Dumazet870c3152014-10-17 09:17:20 -0700932}
David Ahern4297a0e2017-08-07 08:44:21 -0700933
934/* TCP_SKB_CB reference means this can not be used from early demux */
935static inline int tcp_v6_sdif(const struct sk_buff *skb)
936{
937#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
938 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
939 return TCP_SKB_CB(skb)->header.h6.iif;
940#endif
941 return 0;
942}
Eric Dumazetdd2e0b82020-06-19 12:12:35 -0700943
Eric Dumazetb03d2142020-06-23 15:31:12 -0700944extern const struct inet_connection_sock_af_ops ipv6_specific;
945
Eric Dumazetdd2e0b82020-06-19 12:12:35 -0700946INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
Eric Dumazet243600e2020-06-23 15:31:11 -0700947INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
Kuniyuki Iwashima11052582022-07-13 10:52:07 -0700948void tcp_v6_early_demux(struct sk_buff *skb);
Eric Dumazetdd2e0b82020-06-19 12:12:35 -0700949
Eric Dumazet815afe12014-10-18 08:34:37 -0700950#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700951
David Ahern3fa6f612017-08-07 08:44:17 -0700952/* TCP_SKB_CB reference means this can not be used from early demux */
953static inline int tcp_v4_sdif(struct sk_buff *skb)
954{
955#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
956 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
957 return TCP_SKB_CB(skb)->header.h4.iif;
958#endif
959 return 0;
960}
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962/* Due to TSO, an SKB can be composed of multiple actual
963 * packets. To keep these tracked properly, we use this.
964 */
965static inline int tcp_skb_pcount(const struct sk_buff *skb)
966{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700967 return TCP_SKB_CB(skb)->tcp_gso_segs;
968}
969
970static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
971{
972 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
973}
974
975static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
976{
977 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978}
979
Eric Dumazetf69ad292015-06-11 09:15:18 -0700980/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981static inline int tcp_skb_mss(const struct sk_buff *skb)
982{
Eric Dumazetf69ad292015-06-11 09:15:18 -0700983 return TCP_SKB_CB(skb)->tcp_gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984}
985
Martin KaFai Lauc134ecb2016-04-25 14:44:48 -0700986static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
987{
988 return likely(!TCP_SKB_CB(skb)->eor);
989}
990
Mat Martineau85712482020-01-09 07:59:20 -0800991static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
992 const struct sk_buff *from)
993{
994 return likely(tcp_skb_can_collapse_to(to) &&
Talal Ahmad9b65b172021-11-02 22:58:44 -0400995 mptcp_skb_can_collapse(to, from) &&
996 skb_pure_zcopy_same(to, from));
Mat Martineau85712482020-01-09 07:59:20 -0800997}
998
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700999/* Events passed to congestion control interface */
1000enum tcp_ca_event {
1001 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1002 CA_EVENT_CWND_RESTART, /* congestion window restart */
1003 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001004 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +02001005 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
1006 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
Florian Westphal7354c8c2014-09-26 22:37:34 +02001007};
1008
Florian Westphal98900922014-09-26 22:37:35 +02001009/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +02001010enum tcp_ca_ack_event_flags {
Florian Westphalc1d2b4c2017-08-30 19:24:57 +02001011 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1012 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1013 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001014};
1015
1016/*
1017 * Interface for adding new TCP congestion control handlers
1018 */
1019#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -08001020#define TCP_CA_MAX 128
1021#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1022
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001023#define TCP_CA_UNSPEC 0
1024
Daniel Borkmann30e502a2014-09-26 22:37:33 +02001025/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -07001026#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +02001027/* Requires ECN/ECT set on all packets */
1028#define TCP_CONG_NEEDS_ECN 0x2
Martin KaFai Lau0baf26b2020-01-08 16:35:08 -08001029#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
Stephen Hemminger164891a2007-04-23 22:26:16 -07001030
Eric Dumazet64f40ff2015-04-28 16:23:48 -07001031union tcp_cc_info;
1032
Lawrence Brakmo756ee172016-05-11 10:02:13 -07001033struct ack_sample {
1034 u32 pkts_acked;
1035 s32 rtt_us;
Lawrence Brakmo6f094b92016-06-08 21:16:44 -07001036 u32 in_flight;
Lawrence Brakmo756ee172016-05-11 10:02:13 -07001037};
1038
Yuchung Chengb9f64822016-09-19 23:39:14 -04001039/* A rate sample measures the number of (original/retransmitted) data
1040 * packets delivered "delivered" over an interval of time "interval_us".
1041 * The tcp_rate.c code fills in the rate sample, and congestion
1042 * control modules that define a cong_control function to run at the end
1043 * of ACK processing can optionally chose to consult this sample when
1044 * setting cwnd and pacing rate.
1045 * A sample is invalid if "delivered" or "interval_us" is negative.
1046 */
1047struct rate_sample {
Eric Dumazet9a568de2017-05-16 14:00:14 -07001048 u64 prior_mstamp; /* starting timestamp for interval */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001049 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
Yuchung Cheng40bc6062021-09-23 21:17:07 +00001050 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001051 s32 delivered; /* number of packets delivered over interval */
Yuchung Cheng40bc6062021-09-23 21:17:07 +00001052 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
Yuchung Chengb9f64822016-09-19 23:39:14 -04001053 long interval_us; /* time for tp->delivered to incr "delivered" */
Deepti Raghavan4929c942018-07-09 17:53:39 +00001054 u32 snd_interval_us; /* snd interval for delivered packets */
1055 u32 rcv_interval_us; /* rcv interval for delivered packets */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001056 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1057 int losses; /* number of packets marked lost upon ACK */
1058 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1059 u32 prior_in_flight; /* in flight before this ACK */
Pengcheng Yangb253a062022-04-20 10:34:41 +08001060 u32 last_end_seq; /* end_seq of most recently ACKed packet */
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001061 bool is_app_limited; /* is sample from packet with bubble in pipe? */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001062 bool is_retrans; /* is sample from retransmission? */
Yuchung Chenge4286602018-01-17 12:11:01 -08001063 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
Yuchung Chengb9f64822016-09-19 23:39:14 -04001064};
1065
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001066struct tcp_congestion_ops {
Eric Dumazet82506662021-04-02 11:10:37 -07001067/* fast path fields are put first to fill one cache line */
1068
1069 /* return slow start threshold (required) */
1070 u32 (*ssthresh)(struct sock *sk);
1071
1072 /* do new cwnd calculation (required) */
1073 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1074
1075 /* call before changing ca_state (optional) */
1076 void (*set_state)(struct sock *sk, u8 new_state);
1077
1078 /* call when cwnd event occurs (optional) */
1079 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1080
1081 /* call when ack arrives (optional) */
1082 void (*in_ack_event)(struct sock *sk, u32 flags);
1083
1084 /* hook for packet ack accounting (optional) */
1085 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1086
1087 /* override sysctl_tcp_min_tso_segs */
1088 u32 (*min_tso_segs)(struct sock *sk);
1089
1090 /* call when packets are delivered to update cwnd and pacing rate,
1091 * after all the ca_state processing. (optional)
1092 */
1093 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1094
1095
1096 /* new value of cwnd after loss (required) */
1097 u32 (*undo_cwnd)(struct sock *sk);
1098 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1099 u32 (*sndbuf_expand)(struct sock *sk);
1100
1101/* control/slow paths put last */
1102 /* get info for inet_diag (optional) */
1103 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1104 union tcp_cc_info *info);
1105
1106 char name[TCP_CA_NAME_MAX];
1107 struct module *owner;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001108 struct list_head list;
Eric Dumazet82506662021-04-02 11:10:37 -07001109 u32 key;
1110 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001111
1112 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001113 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001114 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001115 void (*release)(struct sock *sk);
Eric Dumazet82506662021-04-02 11:10:37 -07001116} ____cacheline_aligned_in_smp;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001117
Joe Perches5c9f3022013-09-23 11:33:32 -07001118int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1119void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001120
Florian Westphal55d86942014-09-26 22:37:32 +02001121void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001122void tcp_init_congestion_control(struct sock *sk);
1123void tcp_cleanup_congestion_control(struct sock *sk);
Stephen Hemminger6670e152017-11-14 08:25:49 -08001124int tcp_set_default_congestion_control(struct net *net, const char *name);
1125void tcp_get_default_congestion_control(struct net *net, char *name);
Joe Perches5c9f3022013-09-23 11:33:32 -07001126void tcp_get_available_congestion_control(char *buf, size_t len);
1127void tcp_get_allowed_congestion_control(char *buf, size_t len);
1128int tcp_set_allowed_congestion_control(char *allowed);
Eric Dumazet8d650cd2019-07-18 19:28:14 -07001129int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
Neal Cardwell29a94932020-09-10 15:35:34 -04001130 bool cap_net_admin);
Neal Cardwelle73ebb02015-01-28 20:01:35 -05001131u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1132void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001133
Joe Perches5c9f3022013-09-23 11:33:32 -07001134u32 tcp_reno_ssthresh(struct sock *sk);
Florian Westphale9799182016-11-21 14:18:38 +01001135u32 tcp_reno_undo_cwnd(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -07001136void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -07001137extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001138
Martin KaFai Lau0baf26b2020-01-08 16:35:08 -08001139struct tcp_congestion_ops *tcp_ca_find(const char *name);
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001140struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
Stephen Hemminger6670e152017-11-14 08:25:49 -08001141u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
Daniel Borkmannea697632015-01-05 23:57:47 +01001142#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001143char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +01001144#else
1145static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1146{
1147 return NULL;
1148}
1149#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +01001150
Daniel Borkmann30e502a2014-09-26 22:37:33 +02001151static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1152{
1153 const struct inet_connection_sock *icsk = inet_csk(sk);
1154
1155 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1156}
1157
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001158static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001159{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001160 const struct inet_connection_sock *icsk = inet_csk(sk);
1161
1162 if (icsk->icsk_ca_ops->cwnd_event)
1163 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -07001164}
1165
Ping Gan15fcdf62022-04-06 09:09:56 +08001166/* From tcp_cong.c */
1167void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1168
Yuchung Chengb9f64822016-09-19 23:39:14 -04001169/* From tcp_rate.c */
1170void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1171void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1172 struct rate_sample *rs);
1173void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
Yousuk Seungd4761752017-12-07 13:41:34 -08001174 bool is_sack_reneg, struct rate_sample *rs);
Soheil Hassas Yeganehd7722e82016-09-19 23:39:15 -04001175void tcp_rate_check_app_limited(struct sock *sk);
Yuchung Chengb9f64822016-09-19 23:39:14 -04001176
Pengcheng Yangb253a062022-04-20 10:34:41 +08001177static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1178{
1179 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1180}
1181
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001182/* These functions determine how the current flow behaves in respect of SACK
1183 * handling. SACK is negotiated with the peer, and therefore it can vary
1184 * between different flows.
1185 *
1186 * tcp_is_sack - SACK enabled
1187 * tcp_is_reno - No SACK
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001188 */
1189static inline int tcp_is_sack(const struct tcp_sock *tp)
1190{
Eric Dumazetebeef4b2018-11-27 14:42:00 -08001191 return likely(tp->rx_opt.sack_ok);
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001192}
1193
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001194static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +03001195{
1196 return !tcp_is_sack(tp);
1197}
1198
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001199static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1200{
1201 return tp->sacked_out + tp->lost_out;
1202}
1203
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204/* This determines how many packets are "in the network" to the best
1205 * of our knowledge. In many cases it is conservative, but where
1206 * detailed information is available from the receiver (via SACK
1207 * blocks etc.) we can make more aggressive calculations.
1208 *
1209 * Use this for decisions involving congestion control, use just
1210 * tp->packets_out to determine if the send queue is empty or not.
1211 *
1212 * Read this equation as:
1213 *
1214 * "Packets sent once on transmission queue" MINUS
1215 * "Packets left network, but not honestly ACKed yet" PLUS
1216 * "Packets fast retransmitted"
1217 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001218static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +03001220 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221}
1222
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001223#define TCP_INFINITE_SSTHRESH 0x7fffffff
1224
Eric Dumazet40570372022-04-05 16:35:38 -07001225static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1226{
1227 return tp->snd_cwnd;
1228}
1229
1230static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1231{
1232 WARN_ON_ONCE((int)val <= 0);
1233 tp->snd_cwnd = val;
1234}
1235
Yuchung Cheng071d5082015-07-09 13:16:29 -07001236static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1237{
Eric Dumazet40570372022-04-05 16:35:38 -07001238 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
Yuchung Cheng071d5082015-07-09 13:16:29 -07001239}
1240
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -07001241static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1242{
1243 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1244}
1245
Yuchung Cheng684bad12012-09-02 17:38:04 +00001246static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1247{
1248 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1249 (1 << inet_csk(sk)->icsk_ca_state);
1250}
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +00001253 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 * ssthresh.
1255 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001256static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -03001258 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001259
Yuchung Cheng684bad12012-09-02 17:38:04 +00001260 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 return tp->snd_ssthresh;
1262 else
1263 return max(tp->snd_ssthresh,
Eric Dumazet40570372022-04-05 16:35:38 -07001264 ((tcp_snd_cwnd(tp) >> 1) +
1265 (tcp_snd_cwnd(tp) >> 2)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266}
1267
Ilpo Järvinenb9c45952007-07-27 16:36:17 +03001268/* Use define here intentionally to get WARN_ON location shown at the caller */
1269#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
Christoph Paasch5ee2c942014-07-14 16:58:32 +02001271void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -07001272__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Neal Cardwell6b5a5c02011-11-21 17:15:14 +00001274/* The maximum number of MSS of available cwnd for which TSO defers
1275 * sending if not using sysctl_tcp_tso_win_divisor.
1276 */
1277static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1278{
1279 return 3;
1280}
1281
Ilpo Järvinen90840defa2007-12-31 04:48:41 -08001282/* Returns end sequence number of the receiver's advertised window */
1283static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1284{
1285 return tp->snd_una + tp->snd_wnd;
1286}
Eric Dumazete114a712014-04-30 11:58:13 -07001287
1288/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1289 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001290 * it was fully used previously. And that's exactly what we do in
1291 * congestion avoidance mode. But in slow start we allow cwnd to grow
1292 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001293 * Example :
1294 * cwnd is 10 (IW10), but application sends 9 frames.
1295 * We allow cwnd to reach 18 when all frames are ACKed.
1296 * This check is safe because it's as aggressive as slow start which already
1297 * risks 100% overshoot. The advantage is that we discourage application to
1298 * either send more filler packets or data to artificially blow up the cwnd
1299 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001300 */
Eric Dumazet24901552014-05-02 21:18:05 -07001301static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001302{
1303 const struct tcp_sock *tp = tcp_sk(sk);
1304
Neal Cardwellf4ce91c2022-09-28 16:03:31 -04001305 if (tp->is_cwnd_limited)
1306 return true;
1307
Neal Cardwellca8a2262014-05-22 10:41:08 -04001308 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
Yuchung Cheng071d5082015-07-09 13:16:29 -07001309 if (tcp_in_slow_start(tp))
Eric Dumazet40570372022-04-05 16:35:38 -07001310 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
Neal Cardwellca8a2262014-05-22 10:41:08 -04001311
Neal Cardwellf4ce91c2022-09-28 16:03:31 -04001312 return false;
Eric Dumazete114a712014-04-30 11:58:13 -07001313}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001314
Eric Dumazetcadefe52018-06-20 16:07:35 -04001315/* BBR congestion control needs pacing.
1316 * Same remark for SO_MAX_PACING_RATE.
1317 * sch_fq packet scheduler is efficiently handling pacing,
1318 * but is not always installed/used.
1319 * Return true if TCP stack should pace packets itself.
1320 */
1321static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1322{
1323 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1324}
1325
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001326/* Estimates in how many jiffies next packet for this flow can be sent.
1327 * Scheduling a retransmit timer too early would be silly.
Eric Dumazet3f80e082018-10-23 11:54:16 -07001328 */
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001329static inline unsigned long tcp_pacing_delay(const struct sock *sk)
Eric Dumazet3f80e082018-10-23 11:54:16 -07001330{
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001331 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
Eric Dumazet3f80e082018-10-23 11:54:16 -07001332
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001333 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
Eric Dumazet3f80e082018-10-23 11:54:16 -07001334}
1335
1336static inline void tcp_reset_xmit_timer(struct sock *sk,
1337 const int what,
1338 unsigned long when,
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001339 const unsigned long max_when)
Eric Dumazet3f80e082018-10-23 11:54:16 -07001340{
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001341 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
Eric Dumazet3f80e082018-10-23 11:54:16 -07001342 max_when);
1343}
1344
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001345/* Something is really bad, we could not queue an additional packet,
Eric Dumazet3f80e082018-10-23 11:54:16 -07001346 * because qdisc is full or receiver sent a 0 window, or we are paced.
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001347 * We do not want to add fuel to the fire, or abort too early,
1348 * so make sure the timer we arm now is at least 200ms in the future,
1349 * regardless of current icsk_rto value (as it could be ~2ms)
1350 */
1351static inline unsigned long tcp_probe0_base(const struct sock *sk)
1352{
1353 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1354}
1355
1356/* Variant of inet_csk_rto_backoff() used for zero window probes */
1357static inline unsigned long tcp_probe0_when(const struct sock *sk,
1358 unsigned long max_when)
1359{
Cambda Zhu6d4634d2020-12-08 17:19:10 +08001360 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1361 inet_csk(sk)->icsk_backoff);
1362 u64 when = (u64)tcp_probe0_base(sk) << backoff;
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001363
1364 return (unsigned long)min_t(u64, when, max_when);
1365}
1366
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001367static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368{
Eric Dumazet21c8fe92015-05-06 14:26:24 -07001369 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
Eric Dumazet3f80e082018-10-23 11:54:16 -07001370 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
Eric Dumazet8dc242ad2020-05-04 11:27:49 -07001371 tcp_probe0_base(sk), TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372}
1373
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001374static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375{
1376 tp->snd_wl1 = seq;
1377}
1378
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001379static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380{
1381 tp->snd_wl1 = seq;
1382}
1383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384/*
1385 * Calculate(/check) TCP checksum
1386 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001387static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1388 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
Daniel T. Lee0b13c9b2019-04-21 00:50:42 +09001390 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391}
1392
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001393static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394{
Herbert Xu60476372007-04-09 11:59:39 -07001395 return !skb_csum_unnecessary(skb) &&
Cong Wang6ab6dfa2018-11-12 16:17:00 -08001396 __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397}
1398
Menglong Dong7a26dc92022-02-20 15:06:33 +08001399bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1400 enum skb_drop_reason *reason);
Eric Dumazetf35f8212021-11-15 11:02:46 -08001401
Eric Dumazetf35f8212021-11-15 11:02:46 -08001402
Eric Dumazetac6e7802016-11-10 13:12:35 -08001403int tcp_filter(struct sock *sk, struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001404void tcp_set_state(struct sock *sk, int state);
Joe Perches5c9f3022013-09-23 11:33:32 -07001405void tcp_done(struct sock *sk);
Lorenzo Colittic1e64e22015-12-16 12:30:05 +09001406int tcp_abort(struct sock *sk, int err);
1407
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001408static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
1410 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 rx_opt->num_sacks = 0;
1412}
1413
Eric Dumazet6f021c62015-08-21 12:30:00 -07001414void tcp_cwnd_restart(struct sock *sk, s32 delta);
1415
1416static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1417{
Wei Wang1b1fc3f2017-05-05 12:53:23 -07001418 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
Eric Dumazet6f021c62015-08-21 12:30:00 -07001419 struct tcp_sock *tp = tcp_sk(sk);
1420 s32 delta;
1421
Kuniyuki Iwashima4845b572022-07-18 10:26:48 -07001422 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1423 tp->packets_out || ca_ops->cong_control)
Eric Dumazet6f021c62015-08-21 12:30:00 -07001424 return;
Eric Dumazetd635fbe2017-05-16 14:00:03 -07001425 delta = tcp_jiffies32 - tp->lsndtime;
Eric Dumazet6f021c62015-08-21 12:30:00 -07001426 if (delta > inet_csk(sk)->icsk_rto)
1427 tcp_cwnd_restart(sk, delta);
1428}
Yuchung Cheng85f16522013-06-11 15:35:32 -07001429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430/* Determine a window scaling and initial window to offer. */
Eric Dumazetceef9ab2017-10-27 07:47:24 -07001431void tcp_select_initial_window(const struct sock *sk, int __space,
1432 __u32 mss, __u32 *rcv_wnd,
Joe Perches5c9f3022013-09-23 11:33:32 -07001433 __u32 *window_clamp, int wscale_ok,
1434 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Eric Dumazet94f08932017-10-26 21:55:09 -07001436static inline int tcp_win_from_space(const struct sock *sk, int space)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Kuniyuki Iwashima36eeee72022-07-20 09:50:14 -07001438 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
Gao Fengc4836742017-03-24 07:05:12 +08001439
1440 return tcp_adv_win_scale <= 0 ?
1441 (space>>(-tcp_adv_win_scale)) :
1442 space - (space>>tcp_adv_win_scale);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001445/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446static inline int tcp_space(const struct sock *sk)
1447{
Eric Dumazetebb3b782019-10-10 20:17:44 -07001448 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
Eric Dumazet70c26552019-10-09 15:41:03 -07001449 READ_ONCE(sk->sk_backlog.len) -
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001451}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453static inline int tcp_full_space(const struct sock *sk)
1454{
Eric Dumazetebb3b782019-10-10 20:17:44 -07001455 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
Wei Wang053f3682021-09-29 10:25:13 -07001458static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1459{
1460 int unused_mem = sk_unused_reserved_mem(sk);
1461 struct tcp_sock *tp = tcp_sk(sk);
1462
1463 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1464 if (unused_mem)
1465 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1466 tcp_win_from_space(sk, unused_mem));
1467}
1468
Paolo Abenic76c6952020-09-14 10:01:18 +02001469void tcp_cleanup_rbuf(struct sock *sk, int copied);
1470
Eric Dumazet24adbc12020-05-12 06:54:30 -07001471/* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1472 * If 87.5 % (7/8) of the space has been consumed, we want to override
1473 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1474 * len/truesize ratio.
1475 */
1476static inline bool tcp_rmem_pressure(const struct sock *sk)
1477{
Eric Dumazetf969dc52021-02-12 15:22:13 -08001478 int rcvbuf, threshold;
1479
1480 if (tcp_under_memory_pressure(sk))
1481 return true;
1482
1483 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1484 threshold = rcvbuf - (rcvbuf >> 3);
Eric Dumazet24adbc12020-05-12 06:54:30 -07001485
1486 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1487}
1488
Eric Dumazet05dc72a2021-02-12 15:22:14 -08001489static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1490{
1491 const struct tcp_sock *tp = tcp_sk(sk);
1492 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1493
1494 if (avail <= 0)
1495 return false;
1496
1497 return (avail >= target) || tcp_rmem_pressure(sk) ||
1498 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1499}
1500
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001501extern void tcp_openreq_init_rwin(struct request_sock *req,
Eric Dumazetb1964b52015-09-25 07:39:09 -07001502 const struct sock *sk_listener,
1503 const struct dst_entry *dst);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001504
Joe Perches5c9f3022013-09-23 11:33:32 -07001505void tcp_enter_memory_pressure(struct sock *sk);
Eric Dumazet06044752017-06-07 13:29:12 -07001506void tcp_leave_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1509{
Nikolay Borisovb840d152016-01-07 16:38:45 +02001510 struct net *net = sock_net((struct sock *)tp);
1511
Kuniyuki Iwashimaf2f316e2022-07-15 10:17:45 -07001512 return tp->keepalive_intvl ? :
1513 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514}
1515
1516static inline int keepalive_time_when(const struct tcp_sock *tp)
1517{
Nikolay Borisov13b287e2016-01-07 16:38:43 +02001518 struct net *net = sock_net((struct sock *)tp);
1519
Kuniyuki Iwashimaf2f316e2022-07-15 10:17:45 -07001520 return tp->keepalive_time ? :
1521 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522}
1523
Eric Dumazetdf19a622009-08-28 23:48:54 -07001524static inline int keepalive_probes(const struct tcp_sock *tp)
1525{
Nikolay Borisov9bd68612016-01-07 16:38:44 +02001526 struct net *net = sock_net((struct sock *)tp);
1527
Kuniyuki Iwashimaf2f316e2022-07-15 10:17:45 -07001528 return tp->keepalive_probes ? :
1529 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
Eric Dumazetdf19a622009-08-28 23:48:54 -07001530}
1531
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001532static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1533{
1534 const struct inet_connection_sock *icsk = &tp->inet_conn;
1535
Eric Dumazet70eabf02017-05-16 14:00:07 -07001536 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1537 tcp_jiffies32 - tp->rcv_tstamp);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001538}
1539
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001540static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
Kuniyuki Iwashima39e244352022-07-15 10:17:50 -07001542 int fin_timeout = tcp_sk(sk)->linger2 ? :
1543 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001544 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001546 if (fin_timeout < (rto << 2) - (rto >> 1))
1547 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 return fin_timeout;
1550}
1551
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001552static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1553 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001555 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001556 return true;
Arnd Bergmanncca9bab2018-07-11 12:16:12 +02001557 if (unlikely(!time_before32(ktime_get_seconds(),
1558 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001559 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001560 /*
1561 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1562 * then following tcp messages have valid values. Ignore 0 value,
1563 * or else 'negative' tsval might forbid us to accept their packets.
1564 */
1565 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001566 return true;
1567 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001568}
1569
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001570static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1571 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001572{
1573 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001574 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 /* RST segments are not recommended to carry timestamp,
1577 and, if they do, it is recommended to ignore PAWS because
1578 "their cleanup function should take precedence over timestamps."
1579 Certainly, it is mistake. It is necessary to understand the reasons
1580 of this constraint to relax it: if peer reboots, clock may go
1581 out-of-sync and half-open connections will not be reset.
1582 Actually, the problem would be not existing if all
1583 the implementations followed draft about maintaining clock
1584 via reboots. Linux-2.2 DOES NOT!
1585
1586 However, we can relax time bounds for RST segments to MSL.
1587 */
Arnd Bergmanncca9bab2018-07-11 12:16:12 +02001588 if (rst && !time_before32(ktime_get_seconds(),
1589 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001590 return false;
1591 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592}
1593
Eric Dumazet7970ddc2015-03-16 21:06:20 -07001594bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1595 int mib_idx, u32 *last_oow_ack_time);
Neal Cardwell032ee422015-02-06 16:04:38 -05001596
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001597static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598{
1599 /* See RFC 2012 */
Eric Dumazet6aef70a2016-04-27 16:44:27 -07001600 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1601 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1602 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1603 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604}
1605
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001606/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001607static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001608{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001609 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001610}
1611
1612static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1613{
1614 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001615 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001616}
1617
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001618union tcp_md5_addr {
1619 struct in_addr a4;
1620#if IS_ENABLED(CONFIG_IPV6)
1621 struct in6_addr a6;
1622#endif
1623};
1624
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001625/* - key database */
1626struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001627 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001628 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001629 u8 family; /* AF_INET or AF_INET6 */
Ivan Delalande67973182017-06-15 18:07:06 -07001630 u8 prefixlen;
Leonard Cresteza76c2312021-10-15 10:26:05 +03001631 u8 flags;
David Aherndea53bb2019-12-30 14:14:28 -08001632 union tcp_md5_addr addr;
1633 int l3index; /* set if key added with L3 scope */
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001634 u8 key[TCP_MD5SIG_MAXKEYLEN];
1635 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001636};
1637
1638/* - sock block */
1639struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001640 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001641 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001642};
1643
1644/* - pseudo header */
1645struct tcp4_pseudohdr {
1646 __be32 saddr;
1647 __be32 daddr;
1648 __u8 pad;
1649 __u8 protocol;
1650 __be16 len;
1651};
1652
1653struct tcp6_pseudohdr {
1654 struct in6_addr saddr;
1655 struct in6_addr daddr;
1656 __be32 len;
1657 __be32 protocol; /* including padding */
1658};
1659
1660union tcp_md5sum_block {
1661 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001662#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001663 struct tcp6_pseudohdr ip6;
1664#endif
1665};
1666
1667/* - pool: digest algorithm, hash description and scratch buffer */
1668struct tcp_md5sig_pool {
Herbert Xucf80e0e2016-01-24 21:20:23 +08001669 struct ahash_request *md5_req;
Eric Dumazet19689e32016-06-27 18:51:53 +02001670 void *scratch;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001671};
1672
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001673/* - functions */
Eric Dumazet39f8e582015-03-24 15:58:55 -07001674int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1675 const struct sock *sk, const struct sk_buff *skb);
Joe Perches5c9f3022013-09-23 11:33:32 -07001676int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
Leonard Cresteza76c2312021-10-15 10:26:05 +03001677 int family, u8 prefixlen, int l3index, u8 flags,
Dmitry Safonov459837b2022-11-23 17:38:57 +00001678 const u8 *newkey, u8 newkeylen);
1679int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1680 int family, u8 prefixlen, int l3index,
1681 struct tcp_md5sig_key *key);
1682
Joe Perches5c9f3022013-09-23 11:33:32 -07001683int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
Leonard Cresteza76c2312021-10-15 10:26:05 +03001684 int family, u8 prefixlen, int l3index, u8 flags);
Eric Dumazetb83e3de2015-09-25 07:39:15 -07001685struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07001686 const struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001687
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001688#ifdef CONFIG_TCP_MD5SIG
Eric Dumazet6015c712018-11-27 15:03:21 -08001689#include <linux/jump_label.h>
Dmitry Safonov459837b2022-11-23 17:38:57 +00001690extern struct static_key_false_deferred tcp_md5_needed;
David Aherndea53bb2019-12-30 14:14:28 -08001691struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
Eric Dumazet6015c712018-11-27 15:03:21 -08001692 const union tcp_md5_addr *addr,
1693 int family);
1694static inline struct tcp_md5sig_key *
David Aherndea53bb2019-12-30 14:14:28 -08001695tcp_md5_do_lookup(const struct sock *sk, int l3index,
1696 const union tcp_md5_addr *addr, int family)
Eric Dumazet6015c712018-11-27 15:03:21 -08001697{
Dmitry Safonov459837b2022-11-23 17:38:57 +00001698 if (!static_branch_unlikely(&tcp_md5_needed.key))
Eric Dumazet6015c712018-11-27 15:03:21 -08001699 return NULL;
David Aherndea53bb2019-12-30 14:14:28 -08001700 return __tcp_md5_do_lookup(sk, l3index, addr, family);
Eric Dumazet6015c712018-11-27 15:03:21 -08001701}
Jakub Kicinski1330b6e2022-03-07 16:44:21 -08001702
1703enum skb_drop_reason
1704tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1705 const void *saddr, const void *daddr,
1706 int family, int dif, int sdif);
Dmitry Safonov7bbb7652022-02-23 17:57:40 +00001707
Eric Dumazet6015c712018-11-27 15:03:21 -08001708
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001709#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001710#else
David Aherndea53bb2019-12-30 14:14:28 -08001711static inline struct tcp_md5sig_key *
1712tcp_md5_do_lookup(const struct sock *sk, int l3index,
1713 const union tcp_md5_addr *addr, int family)
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001714{
1715 return NULL;
1716}
Jakub Kicinski1330b6e2022-03-07 16:44:21 -08001717
1718static inline enum skb_drop_reason
1719tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1720 const void *saddr, const void *daddr,
Vladimir Oltean24055bb2022-03-09 14:20:12 +02001721 int family, int dif, int sdif)
Dmitry Safonov7bbb7652022-02-23 17:57:40 +00001722{
Jakub Kicinski1330b6e2022-03-07 16:44:21 -08001723 return SKB_NOT_DROPPED_YET;
Dmitry Safonov7bbb7652022-02-23 17:57:40 +00001724}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001725#define tcp_twsk_md5_key(twsk) NULL
1726#endif
1727
Joe Perches5c9f3022013-09-23 11:33:32 -07001728bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001729
Joe Perches5c9f3022013-09-23 11:33:32 -07001730struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001731static inline void tcp_put_md5sig_pool(void)
1732{
1733 local_bh_enable();
1734}
Eric Dumazet35790c02010-05-16 00:34:04 -07001735
Joe Perches5c9f3022013-09-23 11:33:32 -07001736int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1737 unsigned int header_len);
1738int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1739 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001740
Jerry Chu10467162012-08-31 12:29:11 +00001741/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001742void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
Yuchung Cheng72685862017-12-12 13:10:40 -08001743 struct tcp_fastopen_cookie *cookie);
Joe Perches5c9f3022013-09-23 11:33:32 -07001744void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
Daniel Lee2646c832015-04-06 14:37:27 -07001745 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1746 u16 try_exp);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001747struct tcp_fastopen_request {
1748 /* Fast Open cookie. Size 0 means a cookie request */
1749 struct tcp_fastopen_cookie cookie;
1750 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001751 size_t size;
1752 int copied; /* queued in tcp_connect() */
Willem de Bruijnf859a442019-01-25 11:17:23 -05001753 struct ubuf_info *uarg;
Yuchung Cheng783237e2012-07-19 06:43:07 +00001754};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001755void tcp_free_fastopen_req(struct tcp_sock *tp);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07001756void tcp_fastopen_destroy_cipher(struct sock *sk);
Haishuang Yan43713842017-09-27 11:35:42 +08001757void tcp_fastopen_ctx_destroy(struct net *net);
Yuchung Cheng1fba70e2017-10-18 11:22:51 -07001758int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
Ard Biesheuvel438ac882019-06-19 23:46:28 +02001759 void *primary_key, void *backup_key);
Jason Baronf19008e2020-08-10 13:38:39 -04001760int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1761 u64 *key);
Eric Dumazet61d2bca2016-02-01 21:03:07 -08001762void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
Eric Dumazet7c85af82015-09-24 17:16:05 -07001763struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1764 struct request_sock *req,
Christoph Paasch71c02372017-10-23 13:22:23 -07001765 struct tcp_fastopen_cookie *foc,
1766 const struct dst_entry *dst);
Haishuang Yan43713842017-09-27 11:35:42 +08001767void tcp_fastopen_init_key_once(struct net *net);
Wei Wang065263f2017-01-23 10:59:20 -08001768bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1769 struct tcp_fastopen_cookie *cookie);
Wei Wang19f6d3f2017-01-23 10:59:22 -08001770bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
Ard Biesheuvel438ac882019-06-19 23:46:28 +02001771#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
Jason Baron9092a762019-05-29 12:33:57 -04001772#define TCP_FASTOPEN_KEY_MAX 2
1773#define TCP_FASTOPEN_KEY_BUF_LENGTH \
1774 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
Jerry Chu10467162012-08-31 12:29:11 +00001775
1776/* Fastopen key context */
1777struct tcp_fastopen_context {
Ard Biesheuvel438ac882019-06-19 23:46:28 +02001778 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
Ard Biesheuvelc681eda2019-06-17 10:09:33 +02001779 int num;
1780 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001781};
1782
Wei Wang46c2fa32017-04-20 14:45:47 -07001783void tcp_fastopen_active_disable(struct sock *sk);
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001784bool tcp_fastopen_active_should_disable(struct sock *sk);
1785void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
Yuchung Cheng72685862017-12-12 13:10:40 -08001786void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
Wei Wangcf1ef3f2017-04-20 14:45:46 -07001787
Jason Baron9092a762019-05-29 12:33:57 -04001788/* Caller needs to wrap with rcu_read_(un)lock() */
1789static inline
1790struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1791{
1792 struct tcp_fastopen_context *ctx;
1793
1794 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1795 if (!ctx)
1796 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1797 return ctx;
1798}
1799
1800static inline
1801bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1802 const struct tcp_fastopen_cookie *orig)
1803{
1804 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1805 orig->len == foc->len &&
1806 !memcmp(orig->val, foc->val, foc->len))
1807 return true;
1808 return false;
1809}
1810
1811static inline
1812int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1813{
Ard Biesheuvelc681eda2019-06-17 10:09:33 +02001814 return ctx->num;
Jason Baron9092a762019-05-29 12:33:57 -04001815}
1816
Francis Yan05b055e2016-11-27 23:07:13 -08001817/* Latencies incurred by various limits for a sender. They are
1818 * chronograph-like stats that are mutually exclusive.
1819 */
1820enum tcp_chrono {
1821 TCP_CHRONO_UNSPEC,
1822 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1823 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1824 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1825 __TCP_CHRONO_MAX,
1826};
1827
1828void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1829void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1830
Eric Dumazete2080072017-10-04 12:59:58 -07001831/* This helper is needed, because skb->tcp_tsorted_anchor uses
1832 * the same memory storage than skb->destructor/_skb_refdst
1833 */
1834static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1835{
1836 skb->destructor = NULL;
1837 skb->_skb_refdst = 0UL;
1838}
1839
1840#define tcp_skb_tsorted_save(skb) { \
1841 unsigned long _save = skb->_skb_refdst; \
1842 skb->_skb_refdst = 0UL;
1843
1844#define tcp_skb_tsorted_restore(skb) \
1845 skb->_skb_refdst = _save; \
1846}
1847
Eric Dumazetac3f09b2017-10-05 22:21:22 -07001848void tcp_write_queue_purge(struct sock *sk);
David S. Millerfe067e82007-03-07 12:12:44 -08001849
Eric Dumazet75c119a2017-10-05 22:21:27 -07001850static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1851{
1852 return skb_rb_first(&sk->tcp_rtx_queue);
1853}
1854
Eric Dumazetb6171582019-07-19 11:52:33 -07001855static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1856{
1857 return skb_rb_last(&sk->tcp_rtx_queue);
1858}
1859
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001860static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001861{
David S. Millercd07a8e2008-09-23 00:50:13 -07001862 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001863}
1864
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001865#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001866 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001867
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001868static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001869{
Eric Dumazet75c119a2017-10-05 22:21:27 -07001870 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001871}
1872
David S. Millercd07a8e2008-09-23 00:50:13 -07001873static inline bool tcp_skb_is_last(const struct sock *sk,
1874 const struct sk_buff *skb)
1875{
1876 return skb_queue_is_last(&sk->sk_write_queue, skb);
1877}
1878
Eric Dumazetee2aabd2019-12-12 12:55:30 -08001879/**
1880 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1881 * @sk: socket
1882 *
1883 * Since the write queue can have a temporary empty skb in it,
1884 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1885 */
Eric Dumazet75c119a2017-10-05 22:21:27 -07001886static inline bool tcp_write_queue_empty(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001887{
Eric Dumazetee2aabd2019-12-12 12:55:30 -08001888 const struct tcp_sock *tp = tcp_sk(sk);
1889
1890 return tp->write_seq == tp->snd_nxt;
Eric Dumazet75c119a2017-10-05 22:21:27 -07001891}
1892
1893static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1894{
1895 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1896}
1897
1898static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1899{
1900 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
David S. Millerfe067e82007-03-07 12:12:44 -08001901}
1902
David S. Millerfe067e82007-03-07 12:12:44 -08001903static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1904{
Eric Dumazeta43e0522019-02-26 09:49:10 -08001905 __skb_queue_tail(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001906
1907 /* Queue it, remembering where we must start sending. */
Eric Dumazet50895b92017-11-14 21:02:19 -08001908 if (sk->sk_write_queue.next == skb)
Francis Yan0f872302016-11-27 23:07:14 -08001909 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
David S. Millerfe067e82007-03-07 12:12:44 -08001910}
1911
David S. Miller43f59c82008-09-21 21:28:51 -07001912/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001913static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1914 struct sk_buff *skb,
1915 struct sock *sk)
1916{
David S. Miller43f59c82008-09-21 21:28:51 -07001917 __skb_queue_before(&sk->sk_write_queue, skb, new);
David S. Millerfe067e82007-03-07 12:12:44 -08001918}
1919
1920static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1921{
Eric Dumazet4a269812017-10-11 13:27:29 -07001922 tcp_skb_tsorted_anchor_cleanup(skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001923 __skb_unlink(skb, &sk->sk_write_queue);
1924}
1925
Eric Dumazet75c119a2017-10-05 22:21:27 -07001926void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1927
1928static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001929{
Eric Dumazet75c119a2017-10-05 22:21:27 -07001930 tcp_skb_tsorted_anchor_cleanup(skb);
1931 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1932}
1933
1934static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1935{
1936 list_del(&skb->tcp_tsorted_anchor);
1937 tcp_rtx_queue_unlink(skb, sk);
Talal Ahmad03271f32021-10-29 22:05:41 -04001938 tcp_wmem_free_skb(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001939}
1940
Krishna Kumar12d50c42009-12-08 22:26:13 +00001941static inline void tcp_push_pending_frames(struct sock *sk)
1942{
1943 if (tcp_send_head(sk)) {
1944 struct tcp_sock *tp = tcp_sk(sk);
1945
1946 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1947 }
1948}
1949
Neal Cardwellecb97192012-02-27 17:52:52 -05001950/* Start sequence of the skb just after the highest skb with SACKed
1951 * bit, valid only if sacked_out > 0 or when the caller has ensured
1952 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001953 */
1954static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1955{
1956 if (!tp->sacked_out)
1957 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001958
1959 if (tp->highest_sack == NULL)
1960 return tp->snd_nxt;
1961
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001962 return TCP_SKB_CB(tp->highest_sack)->seq;
1963}
1964
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001965static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1966{
Eric Dumazet50895b92017-11-14 21:02:19 -08001967 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001968}
1969
1970static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1971{
1972 return tcp_sk(sk)->highest_sack;
1973}
1974
1975static inline void tcp_highest_sack_reset(struct sock *sk)
1976{
Eric Dumazet50895b92017-11-14 21:02:19 -08001977 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001978}
1979
Eric Dumazet2b7cda92017-10-30 23:08:20 -07001980/* Called when old skb is about to be deleted and replaced by new skb */
1981static inline void tcp_highest_sack_replace(struct sock *sk,
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001982 struct sk_buff *old,
1983 struct sk_buff *new)
1984{
Eric Dumazet2b7cda92017-10-30 23:08:20 -07001985 if (old == tcp_highest_sack(sk))
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001986 tcp_sk(sk)->highest_sack = new;
1987}
1988
Florian Westphalb1f0a0e2015-12-21 21:29:24 +01001989/* This helper checks if socket has IP_TRANSPARENT set */
1990static inline bool inet_sk_transparent(const struct sock *sk)
1991{
1992 switch (sk->sk_state) {
1993 case TCP_TIME_WAIT:
1994 return inet_twsk(sk)->tw_transparent;
1995 case TCP_NEW_SYN_RECV:
1996 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1997 }
1998 return inet_sk(sk)->transparent;
1999}
2000
Andreas Petlund5aa4b322010-02-18 02:45:45 +00002001/* Determines whether this is a thin stream (which may suffer from
2002 * increased latency). Used to trigger latency-reducing mechanisms.
2003 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00002004static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00002005{
2006 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2007}
2008
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009/* /proc */
2010enum tcp_seq_states {
2011 TCP_SEQ_STATE_LISTENING,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013};
2014
Christoph Hellwig37d849b2018-04-11 09:31:28 +02002015void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2016void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2017void tcp_seq_stop(struct seq_file *seq, void *v);
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002018
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00002020 sa_family_t family;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021};
2022
2023struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07002024 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 enum tcp_seq_states state;
2026 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06002027 int bucket, offset, sbucket, num;
Tom Herberta8b690f2010-06-07 00:43:42 -07002028 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029};
2030
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002031extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08002032extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002033
Joe Perches5c9f3022013-09-23 11:33:32 -07002034void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002035
Eric Dumazet28be6e02013-10-18 10:36:17 -07002036struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07002037 netdev_features_t features);
David Millerd4546c22018-06-24 14:13:49 +09002038struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
Eric Dumazet5521d952020-06-23 15:31:14 -07002039INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2040INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2041INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2042INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
Joe Perches5c9f3022013-09-23 11:33:32 -07002043int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00002044
Joe Perches5c9f3022013-09-23 11:33:32 -07002045void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07002046
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002047static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2048{
Nikolay Borisov4979f2d2016-02-03 09:46:57 +02002049 struct net *net = sock_net((struct sock *)tp);
Kuniyuki Iwashima55be8732022-07-15 10:17:51 -07002050 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002051}
2052
Eric Dumazetd3cd4922020-11-13 07:08:08 -08002053bool tcp_stream_memory_free(const struct sock *sk, int wake);
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07002054
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002055#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07002056int tcp4_proc_init(void);
2057void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002058#endif
2059
Eric Dumazetea3bea32015-09-25 07:39:23 -07002060int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03002061int tcp_conn_request(struct request_sock_ops *rsk_ops,
2062 const struct tcp_request_sock_ops *af_ops,
2063 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03002064
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002065/* TCP af-specific functions */
2066struct tcp_sock_af_ops {
2067#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07002068 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07002069 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07002070 int (*calc_md5_hash)(char *location,
2071 const struct tcp_md5sig_key *md5,
2072 const struct sock *sk,
2073 const struct sk_buff *skb);
2074 int (*md5_parse)(struct sock *sk,
Ivan Delalande8917a772017-06-15 18:07:07 -07002075 int optname,
Christoph Hellwigd4c19c42020-07-23 08:09:05 +02002076 sockptr_t optval,
Eric Dumazet39f8e582015-03-24 15:58:55 -07002077 int optlen);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002078#endif
2079};
2080
2081struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03002082 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002083#ifdef CONFIG_TCP_MD5SIG
Eric Dumazetb83e3de2015-09-25 07:39:15 -07002084 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
Eric Dumazetfd3a1542015-03-24 15:58:56 -07002085 const struct sock *addr_sk);
Eric Dumazet39f8e582015-03-24 15:58:55 -07002086 int (*calc_md5_hash) (char *location,
2087 const struct tcp_md5sig_key *md5,
2088 const struct sock *sk,
2089 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002090#endif
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03002091#ifdef CONFIG_SYN_COOKIES
Eric Dumazet3f684b42015-09-29 07:42:49 -07002092 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03002093 __u16 *mss);
2094#endif
Florian Westphal7ea851d2020-11-30 16:36:30 +01002095 struct dst_entry *(*route_req)(const struct sock *sk,
2096 struct sk_buff *skb,
2097 struct flowi *fl,
2098 struct request_sock *req);
Eric Dumazet84b114b2017-05-05 06:56:54 -07002099 u32 (*init_seq)(const struct sk_buff *skb);
Eric Dumazet5d2ed052017-06-07 10:34:39 -07002100 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
Eric Dumazet0f935db2015-09-25 07:39:21 -07002101 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
Octavian Purdilad6274bd2014-06-25 17:09:58 +03002102 struct flowi *fl, struct request_sock *req,
Eric Dumazetdc6ef6b2015-10-16 13:00:01 -07002103 struct tcp_fastopen_cookie *foc,
Martin KaFai Lau331fca42020-08-20 12:00:52 -07002104 enum tcp_synack_type synack_type,
2105 struct sk_buff *syn_skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08002106};
2107
Mat Martineau35b2c322020-01-09 07:59:21 -08002108extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2109#if IS_ENABLED(CONFIG_IPV6)
2110extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2111#endif
2112
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03002113#ifdef CONFIG_SYN_COOKIES
2114static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07002115 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03002116 __u16 *mss)
2117{
Eric Dumazet3f684b42015-09-29 07:42:49 -07002118 tcp_synq_overflow(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07002119 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
Eric Dumazet3f684b42015-09-29 07:42:49 -07002120 return ops->cookie_init_seq(skb, mss);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03002121}
2122#else
2123static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
Eric Dumazet3f684b42015-09-29 07:42:49 -07002124 const struct sock *sk, struct sk_buff *skb,
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03002125 __u16 *mss)
2126{
2127 return 0;
2128}
2129#endif
2130
Joe Perches5c9f3022013-09-23 11:33:32 -07002131int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00002132
Joe Perches5c9f3022013-09-23 11:33:32 -07002133void tcp_v4_init(void);
2134void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002135
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07002136/* tcp_recovery.c */
Yuchung Chengd716bfd2018-05-16 16:40:13 -07002137void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
Yuchung Cheng6ac06ec2018-05-16 16:40:12 -07002138void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
Yuchung Chengb8fef652018-05-16 16:40:16 -07002139extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2140 u32 reo_wnd);
Pengcheng Yang62d9f1a2021-01-24 13:07:14 +08002141extern bool tcp_rack_mark_lost(struct sock *sk);
Yuchung Cheng1d0833d2017-01-12 22:11:34 -08002142extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
Eric Dumazet9a568de2017-05-16 14:00:14 -07002143 u64 xmit_time);
Yuchung Cheng57dde7f2017-01-12 22:11:33 -08002144extern void tcp_rack_reo_timeout(struct sock *sk);
Priyaranjan Jha1f255692017-11-03 16:38:48 -07002145extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
Yuchung Cheng659a8ad2015-10-16 21:57:46 -07002146
Mubashir Adnan Qureshi1a91bb72022-10-26 13:51:12 +00002147/* tcp_plb.c */
2148
2149/*
2150 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2151 * expects cong_ratio which represents fraction of traffic that experienced
2152 * congestion over a single RTT. In order to avoid floating point operations,
2153 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2154 */
2155#define TCP_PLB_SCALE 8
2156
2157/* State for PLB (Protective Load Balancing) for a single TCP connection. */
2158struct tcp_plb_state {
2159 u8 consec_cong_rounds:5, /* consecutive congested rounds */
2160 unused:3;
2161 u32 pause_until; /* jiffies32 when PLB can resume rerouting */
2162};
2163
2164static inline void tcp_plb_init(const struct sock *sk,
2165 struct tcp_plb_state *plb)
2166{
2167 plb->consec_cong_rounds = 0;
2168 plb->pause_until = 0;
2169}
2170void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2171 const int cong_ratio);
2172void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2173void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2174
Neal Cardwelle1a10ef2017-08-03 09:19:52 -04002175/* At how many usecs into the future should the RTO fire? */
2176static inline s64 tcp_rto_delta_us(const struct sock *sk)
2177{
Eric Dumazet75c119a2017-10-05 22:21:27 -07002178 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
Neal Cardwelle1a10ef2017-08-03 09:19:52 -04002179 u32 rto = inet_csk(sk)->icsk_rto;
Eric Dumazet2fd66ff2018-09-21 08:51:47 -07002180 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
Neal Cardwelle1a10ef2017-08-03 09:19:52 -04002181
2182 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2183}
2184
Cong Wange25f8662014-10-15 14:33:21 -07002185/*
2186 * Save and compile IPv4 options, return a pointer to it
2187 */
Paolo Abeni91ed1e62017-08-03 18:07:06 +02002188static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2189 struct sk_buff *skb)
Cong Wange25f8662014-10-15 14:33:21 -07002190{
2191 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2192 struct ip_options_rcu *dopt = NULL;
2193
Cong Wang461b74c2014-10-15 14:33:22 -07002194 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07002195 int opt_size = sizeof(*dopt) + opt->optlen;
2196
2197 dopt = kmalloc(opt_size, GFP_ATOMIC);
Paolo Abeni91ed1e62017-08-03 18:07:06 +02002198 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
Cong Wange25f8662014-10-15 14:33:21 -07002199 kfree(dopt);
2200 dopt = NULL;
2201 }
2202 }
2203 return dopt;
2204}
2205
Eric Dumazet98781962015-02-03 18:31:53 -08002206/* locally generated TCP pure ACKs have skb->truesize == 2
2207 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2208 * This is much faster than dissecting the packet to find out.
2209 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2210 */
2211static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2212{
2213 return skb->truesize == 2;
2214}
2215
2216static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2217{
2218 skb->truesize = 2;
2219}
2220
Tom Herbert473bd232016-03-07 14:11:05 -08002221static inline int tcp_inq(struct sock *sk)
2222{
2223 struct tcp_sock *tp = tcp_sk(sk);
2224 int answ;
2225
2226 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2227 answ = 0;
2228 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2229 !tp->urg_data ||
2230 before(tp->urg_seq, tp->copied_seq) ||
2231 !before(tp->urg_seq, tp->rcv_nxt)) {
2232
2233 answ = tp->rcv_nxt - tp->copied_seq;
2234
2235 /* Subtract 1, if FIN was received */
2236 if (answ && sock_flag(sk, SOCK_DONE))
2237 answ--;
2238 } else {
2239 answ = tp->urg_seq - tp->copied_seq;
2240 }
2241
2242 return answ;
2243}
2244
Tom Herbert32035582016-08-28 14:43:18 -07002245int tcp_peek_len(struct socket *sock);
2246
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07002247static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2248{
2249 u16 segs_in;
2250
2251 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
Eric Dumazet0307a0b2021-11-15 11:02:42 -08002252
2253 /* We update these fields while other threads might
2254 * read them from tcp_get_info()
2255 */
2256 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07002257 if (skb->len > tcp_hdrlen(skb))
Eric Dumazet0307a0b2021-11-15 11:02:42 -08002258 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
Martin KaFai Laua44d6ea2016-03-14 10:52:15 -07002259}
2260
Eric Dumazet9caad862016-04-01 08:52:20 -07002261/*
2262 * TCP listen path runs lockless.
2263 * We forced "struct sock" to be const qualified to make sure
2264 * we don't modify one of its field by mistake.
2265 * Here, we increment sk_drops which is an atomic_t, so we can safely
2266 * make sock writable again.
2267 */
2268static inline void tcp_listendrop(const struct sock *sk)
2269{
2270 atomic_inc(&((struct sock *)sk)->sk_drops);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -07002271 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
Eric Dumazet9caad862016-04-01 08:52:20 -07002272}
2273
Eric Dumazet218af592017-05-16 04:24:36 -07002274enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2275
Dave Watson734942c2017-06-14 11:37:14 -07002276/*
2277 * Interface for adding Upper Level Protocols over TCP
2278 */
2279
2280#define TCP_ULP_NAME_MAX 16
2281#define TCP_ULP_MAX 128
2282#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2283
2284struct tcp_ulp_ops {
2285 struct list_head list;
2286
2287 /* initialize ulp */
2288 int (*init)(struct sock *sk);
John Fastabend95fa1452019-07-19 10:29:22 -07002289 /* update ulp */
John Fastabend33bfe202020-01-11 06:12:01 +00002290 void (*update)(struct sock *sk, struct proto *p,
2291 void (*write_space)(struct sock *sk));
Dave Watson734942c2017-06-14 11:37:14 -07002292 /* cleanup ulp */
2293 void (*release)(struct sock *sk);
Davide Caratti61723b32019-08-30 12:25:48 +02002294 /* diagnostic */
2295 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2296 size_t (*get_info_size)(const struct sock *sk);
Mat Martineau13230592020-01-09 07:59:18 -08002297 /* clone ulp */
2298 void (*clone)(const struct request_sock *req, struct sock *newsk,
2299 const gfp_t priority);
Dave Watson734942c2017-06-14 11:37:14 -07002300
2301 char name[TCP_ULP_NAME_MAX];
2302 struct module *owner;
2303};
2304int tcp_register_ulp(struct tcp_ulp_ops *type);
2305void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2306int tcp_set_ulp(struct sock *sk, const char *name);
2307void tcp_get_available_ulp(char *buf, size_t len);
2308void tcp_cleanup_ulp(struct sock *sk);
John Fastabend33bfe202020-01-11 06:12:01 +00002309void tcp_update_ulp(struct sock *sk, struct proto *p,
2310 void (*write_space)(struct sock *sk));
Dave Watson734942c2017-06-14 11:37:14 -07002311
Daniel Borkmann037b0b82018-08-16 21:49:06 +02002312#define MODULE_ALIAS_TCP_ULP(name) \
2313 __MODULE_INFO(alias, alias_userspace, name); \
2314 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2315
Cong Wang88759602021-02-23 10:49:26 -08002316#ifdef CONFIG_NET_SOCK_MSG
Daniel Borkmann604326b2018-10-13 02:45:58 +02002317struct sk_msg;
2318struct sk_psock;
2319
Cong Wang88759602021-02-23 10:49:26 -08002320#ifdef CONFIG_BPF_SYSCALL
Lorenz Bauerf7476322020-03-09 11:12:36 +00002321struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
Cong Wang51e01582021-04-06 20:21:11 -07002322int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
Lorenz Bauerf7476322020-03-09 11:12:36 +00002323void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
Cong Wang88759602021-02-23 10:49:26 -08002324#endif /* CONFIG_BPF_SYSCALL */
Lorenz Bauerf7476322020-03-09 11:12:36 +00002325
Pengcheng Yanga351d602022-11-29 18:40:39 +08002326int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2327 struct sk_msg *msg, u32 bytes, int flags);
Lorenz Bauer5da00402020-03-09 11:12:35 +00002328#endif /* CONFIG_NET_SOCK_MSG */
Daniel Borkmann604326b2018-10-13 02:45:58 +02002329
Cong Wang88759602021-02-23 10:49:26 -08002330#if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2331static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2332{
2333}
2334#endif
2335
Martin KaFai Lau0813a842020-08-20 12:01:04 -07002336#ifdef CONFIG_CGROUP_BPF
Martin KaFai Lau0813a842020-08-20 12:01:04 -07002337static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2338 struct sk_buff *skb,
2339 unsigned int end_offset)
2340{
2341 skops->skb = skb;
2342 skops->skb_data_end = skb->data + end_offset;
2343}
2344#else
Martin KaFai Lau0813a842020-08-20 12:01:04 -07002345static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2346 struct sk_buff *skb,
2347 unsigned int end_offset)
2348{
2349}
2350#endif
2351
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002352/* Call BPF_SOCK_OPS program that returns an int. If the return value
2353 * is < 0, then the BPF op failed (for example if the loaded BPF
2354 * program does not support the chosen operation or there is no BPF
2355 * program loaded).
2356 */
2357#ifdef CONFIG_BPF
Lawrence Brakmode525be2018-01-25 16:14:09 -08002358static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002359{
2360 struct bpf_sock_ops_kern sock_ops;
2361 int ret;
2362
Lawrence Brakmob73042b2018-01-25 16:14:08 -08002363 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
Lawrence Brakmof19397a2017-12-01 10:15:04 -08002364 if (sk_fullsock(sk)) {
2365 sock_ops.is_fullsock = 1;
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002366 sock_owned_by_me(sk);
Lawrence Brakmof19397a2017-12-01 10:15:04 -08002367 }
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002368
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002369 sock_ops.sk = sk;
2370 sock_ops.op = op;
Lawrence Brakmode525be2018-01-25 16:14:09 -08002371 if (nargs > 0)
2372 memcpy(sock_ops.args, args, nargs * sizeof(*args));
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002373
2374 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2375 if (ret == 0)
2376 ret = sock_ops.reply;
2377 else
2378 ret = -1;
2379 return ret;
2380}
Lawrence Brakmode525be2018-01-25 16:14:09 -08002381
2382static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2383{
2384 u32 args[2] = {arg1, arg2};
2385
2386 return tcp_call_bpf(sk, op, 2, args);
2387}
2388
2389static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2390 u32 arg3)
2391{
2392 u32 args[3] = {arg1, arg2, arg3};
2393
2394 return tcp_call_bpf(sk, op, 3, args);
2395}
2396
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002397#else
Lawrence Brakmode525be2018-01-25 16:14:09 -08002398static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002399{
2400 return -EPERM;
2401}
Lawrence Brakmode525be2018-01-25 16:14:09 -08002402
2403static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2404{
2405 return -EPERM;
2406}
2407
2408static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2409 u32 arg3)
2410{
2411 return -EPERM;
2412}
2413
Lawrence Brakmo40304b22017-06-30 20:02:40 -07002414#endif
2415
Lawrence Brakmo8550f322017-06-30 20:02:42 -07002416static inline u32 tcp_timeout_init(struct sock *sk)
2417{
2418 int timeout;
2419
Lawrence Brakmode525be2018-01-25 16:14:09 -08002420 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
Lawrence Brakmo8550f322017-06-30 20:02:42 -07002421
2422 if (timeout <= 0)
2423 timeout = TCP_TIMEOUT_INIT;
Akhmat Karakotov59031232022-01-28 22:26:21 +03002424 return min_t(int, timeout, TCP_RTO_MAX);
Lawrence Brakmo8550f322017-06-30 20:02:42 -07002425}
2426
Lawrence Brakmo13d3b1e2017-06-30 20:02:44 -07002427static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2428{
2429 int rwnd;
2430
Lawrence Brakmode525be2018-01-25 16:14:09 -08002431 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
Lawrence Brakmo13d3b1e2017-06-30 20:02:44 -07002432
2433 if (rwnd < 0)
2434 rwnd = 0;
2435 return rwnd;
2436}
Lawrence Brakmo91b5b212017-06-30 20:02:49 -07002437
2438static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2439{
Lawrence Brakmode525be2018-01-25 16:14:09 -08002440 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
Lawrence Brakmo91b5b212017-06-30 20:02:49 -07002441}
Ursula Braun60e2a772017-10-25 11:01:45 +02002442
Stanislav Fomichev23729ff2019-07-02 09:13:56 -07002443static inline void tcp_bpf_rtt(struct sock *sk)
2444{
Arnd Bergmannbef8e262019-07-08 14:57:21 +02002445 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
Stanislav Fomichev23729ff2019-07-02 09:13:56 -07002446 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2447}
2448
Ursula Braun60e2a772017-10-25 11:01:45 +02002449#if IS_ENABLED(CONFIG_SMC)
2450extern struct static_key_false tcp_have_smc;
2451#endif
Ilya Lesokhin6dac1522018-04-30 10:16:10 +03002452
2453#if IS_ENABLED(CONFIG_TLS_DEVICE)
2454void clean_acked_data_enable(struct inet_connection_sock *icsk,
2455 void (*cad)(struct sock *sk, u32 ack_seq));
2456void clean_acked_data_disable(struct inet_connection_sock *icsk);
Jakub Kicinski494bc1d2019-05-08 16:46:14 -07002457void clean_acked_data_flush(void);
Ilya Lesokhin6dac1522018-04-30 10:16:10 +03002458#endif
2459
Eric Dumazeta842fe12019-06-12 11:57:25 -07002460DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2461static inline void tcp_add_tx_delay(struct sk_buff *skb,
2462 const struct tcp_sock *tp)
2463{
2464 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2465 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2466}
2467
Eric Dumazetd6fb3962019-06-13 21:22:35 -07002468/* Compute Earliest Departure Time for some control packets
2469 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2470 */
2471static inline u64 tcp_transmit_time(const struct sock *sk)
Eric Dumazeta842fe12019-06-12 11:57:25 -07002472{
2473 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2474 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2475 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2476
Eric Dumazetd6fb3962019-06-13 21:22:35 -07002477 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
Eric Dumazeta842fe12019-06-12 11:57:25 -07002478 }
Eric Dumazetd6fb3962019-06-13 21:22:35 -07002479 return 0;
Eric Dumazeta842fe12019-06-12 11:57:25 -07002480}
2481
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482#endif /* _TCP_H */