blob: f87599d5af8236148dff21cb4d1ff1c708ed1c8a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#define FASTRETRANS_DEBUG 1
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/list.h>
24#include <linux/tcp.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050025#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
Herbert Xufb286bb2005-11-10 13:01:24 -080029#include <linux/skbuff.h>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080030#include <linux/crypto.h>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080031#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000032#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080033#include <linux/ktime.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070034
35#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070036#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070037#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070039#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <net/sock.h>
41#include <net/snmp.h>
42#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070043#include <net/tcp_states.h>
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -070044#include <net/inet_ecn.h>
Satoru SATOH0c266892009-05-04 11:11:01 -070045#include <net/dst.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/seq_file.h>
Glauber Costa180d8cd2011-12-11 21:47:02 +000048#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070050extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Eric Dumazetdd24c002008-11-25 21:17:14 -080052extern struct percpu_counter tcp_orphan_count;
Joe Perches5c9f3022013-09-23 11:33:32 -070053void tcp_time_wait(struct sock *sk, int state, int timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070056#define MAX_TCP_OPTION_SPACE 40
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090058/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 * Never offer a window over 32767 without using window scaling. Some
Kenjiro Nakayama105970f2014-10-20 18:15:50 +090060 * poor stacks do signed 16bit maths!
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 */
62#define MAX_TCP_WINDOW 32767U
63
64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65#define TCP_MIN_MSS 88U
66
John Heffner5d424d52006-03-20 17:53:41 -080067/* The least MTU to use for probing */
68#define TCP_BASE_MSS 512
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070/* After receiving this amount of duplicate ACKs fast retransmit starts. */
71#define TCP_FASTRETRANS_THRESH 3
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073/* Maximal number of ACKs sent quickly to accelerate slow-start. */
74#define TCP_MAX_QUICKACKS 16U
75
76/* urg_data states */
77#define TCP_URG_VALID 0x0100
78#define TCP_URG_NOTYET 0x0200
79#define TCP_URG_READ 0x0400
80
81#define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
86 */
87
88#define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
93 */
94
Alex Bergmann6c9ff972012-08-31 02:48:31 +000095#define TCP_SYN_RETRIES 6 /* This is how many retries are done
96 * when active opening a connection.
97 * RFC1122 says the minimum retry MUST
98 * be at least 180secs. Nevertheless
99 * this value is corresponding to
100 * 63secs of retransmission with the
101 * current initial RTO.
102 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000104#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
105 * when passive opening a connection.
106 * This is corresponding to 31secs of
107 * retransmission with the current
108 * initial RTO.
109 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
112 * state, about 60 seconds */
113#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
114 /* BSD style FIN_WAIT2 deadlock breaker.
115 * It used to be 3min, new value is 60sec,
116 * to combine FIN-WAIT-2 timeout with
117 * TIME-WAIT timer.
118 */
119
120#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
121#if HZ >= 100
122#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
123#define TCP_ATO_MIN ((unsigned)(HZ/25))
124#else
125#define TCP_DELACK_MIN 4U
126#define TCP_ATO_MIN 4U
127#endif
128#define TCP_RTO_MAX ((unsigned)(120*HZ))
129#define TCP_RTO_MIN ((unsigned)(HZ/5))
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000130#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000131#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
132 * used as a fallback RTO for the
133 * initial data transmission if no
134 * valid RTT sample has been acquired,
135 * most likely due to retrans in 3WHS.
136 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
139 * for local resources.
140 */
141
142#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
143#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
144#define TCP_KEEPALIVE_INTVL (75*HZ)
145
146#define MAX_TCP_KEEPIDLE 32767
147#define MAX_TCP_KEEPINTVL 32767
148#define MAX_TCP_KEEPCNT 127
149#define MAX_TCP_SYNCNT 127
150
151#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
153#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
154#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
155 * after this time. It should be equal
156 * (or greater than) TCP_TIMEWAIT_LEN
157 * to provide reliability equal to one
158 * provided by timewait state.
159 */
160#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
161 * timestamps. It must be less than
162 * minimal timewait lifetime.
163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164/*
165 * TCP option
166 */
Kenjiro Nakayama105970f2014-10-20 18:15:50 +0900167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168#define TCPOPT_NOP 1 /* Padding */
169#define TCPOPT_EOL 0 /* End of options */
170#define TCPOPT_MSS 2 /* Segment size negotiating */
171#define TCPOPT_WINDOW 3 /* Window scaling */
172#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
173#define TCPOPT_SACK 5 /* SACK Block */
174#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800175#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000176#define TCPOPT_EXP 254 /* Experimental */
177/* Magic number to be after the option value for sharing TCP
178 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
179 */
180#define TCPOPT_FASTOPEN_MAGIC 0xF989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
182/*
183 * TCP option lengths
184 */
185
186#define TCPOLEN_MSS 4
187#define TCPOLEN_WINDOW 3
188#define TCPOLEN_SACK_PERM 2
189#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800190#define TCPOLEN_MD5SIG 18
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000191#define TCPOLEN_EXP_FASTOPEN_BASE 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193/* But this is what stacks really send out. */
194#define TCPOLEN_TSTAMP_ALIGNED 12
195#define TCPOLEN_WSCALE_ALIGNED 4
196#define TCPOLEN_SACKPERM_ALIGNED 4
197#define TCPOLEN_SACK_BASE 2
198#define TCPOLEN_SACK_BASE_ALIGNED 4
199#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800200#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700201#define TCPOLEN_MSS_ALIGNED 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203/* Flags in tp->nonagle */
204#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
205#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800206#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000208/* TCP thin-stream limits */
209#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
210
David S. Miller7eb38522011-02-05 18:13:45 -0800211/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
David S. Miller442b9632011-02-02 17:05:11 -0800212#define TCP_INIT_CWND 10
213
Yuchung Chengcf60af02012-07-19 06:43:09 +0000214/* Bit Flags for sysctl_tcp_fastopen */
215#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000216#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000217#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af02012-07-19 06:43:09 +0000218
Jerry Chu10467162012-08-31 12:29:11 +0000219/* Accept SYN data w/o any cookie option */
220#define TFO_SERVER_COOKIE_NOT_REQD 0x200
221
222/* Force enable TFO on all listeners, i.e., not requiring the
223 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
224 */
225#define TFO_SERVER_WO_SOCKOPT1 0x400
226#define TFO_SERVER_WO_SOCKOPT2 0x800
Jerry Chu10467162012-08-31 12:29:11 +0000227
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700228extern struct inet_timewait_death_row tcp_death_row;
229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231extern int sysctl_tcp_timestamps;
232extern int sysctl_tcp_window_scaling;
233extern int sysctl_tcp_sack;
234extern int sysctl_tcp_fin_timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235extern int sysctl_tcp_keepalive_time;
236extern int sysctl_tcp_keepalive_probes;
237extern int sysctl_tcp_keepalive_intvl;
238extern int sysctl_tcp_syn_retries;
239extern int sysctl_tcp_synack_retries;
240extern int sysctl_tcp_retries1;
241extern int sysctl_tcp_retries2;
242extern int sysctl_tcp_orphan_retries;
243extern int sysctl_tcp_syncookies;
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000244extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245extern int sysctl_tcp_retrans_collapse;
246extern int sysctl_tcp_stdurg;
247extern int sysctl_tcp_rfc1337;
248extern int sysctl_tcp_abort_on_overflow;
249extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250extern int sysctl_tcp_fack;
251extern int sysctl_tcp_reordering;
Eric Dumazetdca145f2014-10-27 21:45:24 -0700252extern int sysctl_tcp_max_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700254extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255extern int sysctl_tcp_wmem[3];
256extern int sysctl_tcp_rmem[3];
257extern int sysctl_tcp_app_win;
258extern int sysctl_tcp_adv_win_scale;
259extern int sysctl_tcp_tw_reuse;
260extern int sysctl_tcp_frto;
261extern int sysctl_tcp_low_latency;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263extern int sysctl_tcp_moderate_rcvbuf;
264extern int sysctl_tcp_tso_win_divisor;
Rick Jones15d99e02006-03-20 22:40:29 -0800265extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700266extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000267extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000268extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000269extern int sysctl_tcp_early_retrans;
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000270extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200271extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -0700272extern unsigned int sysctl_tcp_notsent_lowat;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700273extern int sysctl_tcp_min_tso_segs;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800274extern int sysctl_tcp_autocorking;
Neal Cardwell032ee422015-02-06 16:04:38 -0500275extern int sysctl_tcp_invalid_ratelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Eric Dumazet8d987e52010-11-09 23:24:26 +0000277extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800278extern struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279extern int tcp_memory_pressure;
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 * The next routines deal with comparing 32 bit unsigned ints
283 * and worry about wraparound (automatic with unsigned arithmetic).
284 */
285
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000286static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800288 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800290#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
292/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000293static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 return seq3 - seq2 >= seq1 - seq2;
296}
297
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800298static inline bool tcp_out_of_memory(struct sock *sk)
299{
300 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
301 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
302 return true;
303 return false;
304}
305
David S. Millerad1af0f2010-08-25 02:27:49 -0700306static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700307{
David S. Millerad1af0f2010-08-25 02:27:49 -0700308 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
309 int orphans = percpu_counter_read_positive(ocp);
310
311 if (orphans << shift > sysctl_tcp_max_orphans) {
312 orphans = percpu_counter_sum_positive(ocp);
313 if (orphans << shift > sysctl_tcp_max_orphans)
314 return true;
315 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700316 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700317}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
Joe Perches5c9f3022013-09-23 11:33:32 -0700319bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800320
Florian Westphala0f82f62009-04-19 09:43:48 +0000321/* syncookies: remember time of last synqueue overflow */
322static inline void tcp_synq_overflow(struct sock *sk)
323{
324 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
325}
326
327/* syncookies: no recent synqueue overflow on this listening socket? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000328static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
Florian Westphala0f82f62009-04-19 09:43:48 +0000329{
330 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
Jerry Chu9ad7c042011-06-08 11:08:38 +0000331 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
Florian Westphala0f82f62009-04-19 09:43:48 +0000332}
333
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334extern struct proto tcp_prot;
335
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700336#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
337#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
338#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
339#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
Tom Herbertaa2ea052010-04-22 07:00:24 +0000340#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
Joe Perches5c9f3022013-09-23 11:33:32 -0700342void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000343
Joe Perches5c9f3022013-09-23 11:33:32 -0700344void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Joe Perches5c9f3022013-09-23 11:33:32 -0700346void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Joe Perches5c9f3022013-09-23 11:33:32 -0700348void tcp_v4_early_demux(struct sk_buff *skb);
349int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
Joe Perches5c9f3022013-09-23 11:33:32 -0700351int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
Ying Xue1b784142015-03-02 15:37:48 +0800352int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
Joe Perches5c9f3022013-09-23 11:33:32 -0700353int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
354 int flags);
355void tcp_release_cb(struct sock *sk);
356void tcp_wfree(struct sk_buff *skb);
357void tcp_write_timer_handler(struct sock *sk);
358void tcp_delack_timer_handler(struct sock *sk);
359int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
360int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
361 const struct tcphdr *th, unsigned int len);
362void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
363 const struct tcphdr *th, unsigned int len);
364void tcp_rcv_space_adjust(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700365int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
366void tcp_twsk_destructor(struct sock *sk);
367ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
368 struct pipe_inode_info *pipe, size_t len,
369 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800370
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700371static inline void tcp_dec_quickack_mode(struct sock *sk,
372 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700374 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700375
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700376 if (icsk->icsk_ack.quick) {
377 if (pkts >= icsk->icsk_ack.quick) {
378 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700379 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700380 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700381 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700382 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
384}
385
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700386#define TCP_ECN_OK 1
387#define TCP_ECN_QUEUE_CWR 2
388#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000389#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700390
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000391enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 TCP_TW_SUCCESS = 0,
393 TCP_TW_RST = 1,
394 TCP_TW_ACK = 2,
395 TCP_TW_SYN = 3
396};
397
398
Joe Perches5c9f3022013-09-23 11:33:32 -0700399enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
400 struct sk_buff *skb,
401 const struct tcphdr *th);
402struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
403 struct request_sock *req, struct request_sock **prev,
404 bool fastopen);
405int tcp_child_process(struct sock *parent, struct sock *child,
406 struct sk_buff *skb);
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400407void tcp_enter_loss(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700408void tcp_clear_retrans(struct tcp_sock *tp);
409void tcp_update_metrics(struct sock *sk);
410void tcp_init_metrics(struct sock *sk);
411void tcp_metrics_init(void);
412bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
Hannes Frederic Sowaa26552a2014-08-14 22:06:12 +0200413 bool paws_check, bool timestamps);
Joe Perches5c9f3022013-09-23 11:33:32 -0700414bool tcp_remember_stamp(struct sock *sk);
415bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
416void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
417void tcp_disable_fack(struct tcp_sock *tp);
418void tcp_close(struct sock *sk, long timeout);
419void tcp_init_sock(struct sock *sk);
420unsigned int tcp_poll(struct file *file, struct socket *sock,
421 struct poll_table_struct *wait);
422int tcp_getsockopt(struct sock *sk, int level, int optname,
423 char __user *optval, int __user *optlen);
424int tcp_setsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, unsigned int optlen);
426int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000427 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700428int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000429 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700430void tcp_set_keepalive(struct sock *sk, int val);
431void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
Ying Xue1b784142015-03-02 15:37:48 +0800432int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
433 int flags, int *addr_len);
Joe Perches5c9f3022013-09-23 11:33:32 -0700434void tcp_parse_options(const struct sk_buff *skb,
435 struct tcp_options_received *opt_rx,
436 int estab, struct tcp_fastopen_cookie *foc);
437const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439/*
440 * TCP v4 functions exported for the inet6 API
441 */
442
Joe Perches5c9f3022013-09-23 11:33:32 -0700443void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
Neal Cardwell4fab9072014-08-14 12:40:05 -0400444void tcp_v4_mtu_reduced(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700445int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
446struct sock *tcp_create_openreq_child(struct sock *sk,
447 struct request_sock *req,
448 struct sk_buff *skb);
Daniel Borkmann81164412015-01-05 23:57:48 +0100449void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
Joe Perches5c9f3022013-09-23 11:33:32 -0700450struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
451 struct request_sock *req,
452 struct dst_entry *dst);
453int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
454int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
455int tcp_connect(struct sock *sk);
456struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
457 struct request_sock *req,
458 struct tcp_fastopen_cookie *foc);
459int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000461void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000462int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000463void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465/* From syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700466int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
467 u32 cookie);
Cong Wang461b74c2014-10-15 14:33:22 -0700468struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400469#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200470
Eric Dumazet63262312014-03-19 21:02:21 -0700471/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200472 * This counter is used both as a hash input and partially encoded into
473 * the cookie value. A cookie is only validated further if the delta
474 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700475 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200476 * the counter advances immediately after a cookie is generated).
477 */
478#define MAX_SYNCOOKIE_AGE 2
479
480static inline u32 tcp_cookie_time(void)
481{
Eric Dumazet63262312014-03-19 21:02:21 -0700482 u64 val = get_jiffies_64();
483
484 do_div(val, 60 * HZ);
485 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200486}
487
Joe Perches5c9f3022013-09-23 11:33:32 -0700488u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
489 u16 *mssp);
Octavian Purdila57b47552014-06-25 17:09:50 +0300490__u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
491 __u16 *mss);
Joe Perches5c9f3022013-09-23 11:33:32 -0700492__u32 cookie_init_timestamp(struct request_sock *req);
Florian Westphalf1673382014-11-03 17:35:02 +0100493bool cookie_timestamp_decode(struct tcp_options_received *opt);
494bool cookie_ecn_ok(const struct tcp_options_received *opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100495 const struct net *net, const struct dst_entry *dst);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700496
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800497/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700498int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
499 u32 cookie);
500struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Florian Westphalf1673382014-11-03 17:35:02 +0100501
Joe Perches5c9f3022013-09-23 11:33:32 -0700502u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
503 const struct tcphdr *th, u16 *mssp);
504__u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
505 __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400506#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507/* tcp_output.c */
508
Joe Perches5c9f3022013-09-23 11:33:32 -0700509void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
510 int nonagle);
511bool tcp_may_send_now(struct sock *sk);
512int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
513int tcp_retransmit_skb(struct sock *, struct sk_buff *);
514void tcp_retransmit_timer(struct sock *sk);
515void tcp_xmit_retransmit_queue(struct sock *);
516void tcp_simple_retransmit(struct sock *);
517int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300518int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Joe Perches5c9f3022013-09-23 11:33:32 -0700520void tcp_send_probe0(struct sock *);
521void tcp_send_partial(struct sock *);
522int tcp_write_wakeup(struct sock *);
523void tcp_send_fin(struct sock *sk);
524void tcp_send_active_reset(struct sock *sk, gfp_t priority);
525int tcp_send_synack(struct sock *);
526bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
527 const char *proto);
528void tcp_push_one(struct sock *, unsigned int mss_now);
529void tcp_send_ack(struct sock *sk);
530void tcp_send_delayed_ack(struct sock *sk);
531void tcp_send_loss_probe(struct sock *sk);
532bool tcp_schedule_loss_probe(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
David S. Millera762a982005-07-05 15:18:51 -0700534/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700535void tcp_resume_early_retransmit(struct sock *sk);
536void tcp_rearm_rto(struct sock *sk);
537void tcp_reset(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700540void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700541static inline void tcp_clear_xmit_timers(struct sock *sk)
542{
543 inet_csk_clear_xmit_timers(sk);
544}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Joe Perches5c9f3022013-09-23 11:33:32 -0700546unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
547unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000548
549/* Bound MSS / TSO packet size with the half of the window */
550static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
551{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700552 int cutoff;
553
554 /* When peer uses tiny windows, there is no use in packetizing
555 * to sub-MSS pieces for the sake of SWS or making sure there
556 * are enough packets in the pipe for fast recovery.
557 *
558 * On the other hand, for extremely large MSS devices, handling
559 * smaller than MSS windows in this way does make sense.
560 */
561 if (tp->max_window >= 512)
562 cutoff = (tp->max_window >> 1);
563 else
564 cutoff = tp->max_window;
565
566 if (cutoff && pktsize > cutoff)
567 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000568 else
569 return pktsize;
570}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300572/* tcp.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700573void tcp_get_info(const struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575/* Read 'sendfile()'-style from a TCP socket */
576typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
577 unsigned int, size_t);
Joe Perches5c9f3022013-09-23 11:33:32 -0700578int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
579 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Joe Perches5c9f3022013-09-23 11:33:32 -0700581void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Joe Perches5c9f3022013-09-23 11:33:32 -0700583int tcp_mtu_to_mss(struct sock *sk, int pmtu);
584int tcp_mss_to_mtu(struct sock *sk, int mss);
585void tcp_mtup_init(struct sock *sk);
586void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800587
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000588static inline void tcp_bound_rto(const struct sock *sk)
589{
590 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
591 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
592}
593
594static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
595{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800596 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000597}
598
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800599static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
601 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
602 ntohl(TCP_FLAG_ACK) |
603 snd_wnd);
604}
605
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800606static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
608 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
609}
610
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700611static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700613 struct tcp_sock *tp = tcp_sk(sk);
614
David S. Millerb03efcf2005-07-08 14:57:23 -0700615 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 tp->rcv_wnd &&
617 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
618 !tp->urg_data)
619 tcp_fast_path_on(tp);
620}
621
Satoru SATOH0c266892009-05-04 11:11:01 -0700622/* Compute the actual rto_min value */
623static inline u32 tcp_rto_min(struct sock *sk)
624{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400625 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700626 u32 rto_min = TCP_RTO_MIN;
627
628 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
629 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
630 return rto_min;
631}
632
Eric Dumazet740b0f12014-02-26 14:02:48 -0800633static inline u32 tcp_rto_min_us(struct sock *sk)
634{
635 return jiffies_to_usecs(tcp_rto_min(sk));
636}
637
Daniel Borkmann81164412015-01-05 23:57:48 +0100638static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
639{
640 return dst_metric_locked(dst, RTAX_CC_ALGO);
641}
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643/* Compute the actual receive window we are currently advertising.
644 * Rcv_nxt can be after the window if our peer push more data
645 * than the offered window.
646 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800647static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
650
651 if (win < 0)
652 win = 0;
653 return (u32) win;
654}
655
656/* Choose a new window, without checks for shrinking, and without
657 * scaling applied to the result. The caller does these things
658 * if necessary. This is a "raw" window selection.
659 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700660u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Pavel Emelyanovee995282012-04-19 03:40:39 +0000662void tcp_send_window_probe(struct sock *sk);
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664/* TCP timestamps are only 32-bits, this causes a slight
665 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800666 * of jiffies in the buffer control blocks below. We decided
667 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 * casts with the following macro.
669 */
670#define tcp_time_stamp ((__u32)(jiffies))
671
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700672static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
673{
674 return skb->skb_mstamp.stamp_jiffies;
675}
676
677
Changli Gaoa3433f32010-06-12 14:01:43 +0000678#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
679
680#define TCPHDR_FIN 0x01
681#define TCPHDR_SYN 0x02
682#define TCPHDR_RST 0x04
683#define TCPHDR_PSH 0x08
684#define TCPHDR_ACK 0x10
685#define TCPHDR_URG 0x20
686#define TCPHDR_ECE 0x40
687#define TCPHDR_CWR 0x80
688
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800689/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700690 * TCP per-packet control information to the transmission code.
691 * We also store the host-order sequence numbers in here too.
692 * This is 44 bytes if IPV6 is enabled.
693 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 */
695struct tcp_skb_cb {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 __u32 seq; /* Starting sequence number */
697 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700698 union {
699 /* Note : tcp_tw_isn is used in input path only
700 * (isn chosen by tcp_timewait_state_process())
701 *
702 * tcp_gso_segs is used in write queue only,
703 * cf tcp_skb_pcount()
704 */
705 __u32 tcp_tw_isn;
706 __u32 tcp_gso_segs;
707 };
Eric Dumazet4de075e2011-09-27 13:25:05 -0400708 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 __u8 sacked; /* State flags for SACK/FACK. */
711#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
712#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
713#define TCPCB_LOST 0x04 /* SKB is lost */
714#define TCPCB_TAGBITS 0x07 /* All tag bits */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400715#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
Andrey Vagin9d186ca2014-08-13 16:03:10 +0400717#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
718 TCPCB_REPAIRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000720 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
721 /* 1 byte hole */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 __u32 ack_seq; /* Sequence number ACK'd */
Eric Dumazet971f10e2014-09-27 09:50:57 -0700723 union {
724 struct inet_skb_parm h4;
725#if IS_ENABLED(CONFIG_IPV6)
726 struct inet6_skb_parm h6;
727#endif
728 } header; /* For incoming frames */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729};
730
731#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
732
Eric Dumazet870c3152014-10-17 09:17:20 -0700733
Eric Dumazet815afe12014-10-18 08:34:37 -0700734#if IS_ENABLED(CONFIG_IPV6)
Eric Dumazet870c3152014-10-17 09:17:20 -0700735/* This is the variant of inet6_iif() that must be used by TCP,
736 * as TCP moves IP6CB into a different location in skb->cb[]
737 */
738static inline int tcp_v6_iif(const struct sk_buff *skb)
739{
740 return TCP_SKB_CB(skb)->header.h6.iif;
741}
Eric Dumazet815afe12014-10-18 08:34:37 -0700742#endif
Eric Dumazet870c3152014-10-17 09:17:20 -0700743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744/* Due to TSO, an SKB can be composed of multiple actual
745 * packets. To keep these tracked properly, we use this.
746 */
747static inline int tcp_skb_pcount(const struct sk_buff *skb)
748{
Eric Dumazetcd7d8492014-09-24 04:11:22 -0700749 return TCP_SKB_CB(skb)->tcp_gso_segs;
750}
751
752static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
753{
754 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
755}
756
757static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
758{
759 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}
761
762/* This is valid iff tcp_skb_pcount() > 1. */
763static inline int tcp_skb_mss(const struct sk_buff *skb)
764{
Herbert Xu79671682006-06-22 02:40:14 -0700765 return skb_shinfo(skb)->gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700768/* Events passed to congestion control interface */
769enum tcp_ca_event {
770 CA_EVENT_TX_START, /* first transmit when no packets in flight */
771 CA_EVENT_CWND_RESTART, /* congestion window restart */
772 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700773 CA_EVENT_LOSS, /* loss timeout */
Florian Westphal98900922014-09-26 22:37:35 +0200774 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
775 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
776 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
777 CA_EVENT_NON_DELAYED_ACK,
Florian Westphal7354c8c2014-09-26 22:37:34 +0200778};
779
Florian Westphal98900922014-09-26 22:37:35 +0200780/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
Florian Westphal7354c8c2014-09-26 22:37:34 +0200781enum tcp_ca_ack_event_flags {
Florian Westphal98900922014-09-26 22:37:35 +0200782 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
783 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
784 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700785};
786
787/*
788 * Interface for adding new TCP congestion control handlers
789 */
790#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800791#define TCP_CA_MAX 128
792#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
793
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100794#define TCP_CA_UNSPEC 0
795
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200796/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
Stephen Hemminger164891a2007-04-23 22:26:16 -0700797#define TCP_CONG_NON_RESTRICTED 0x1
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200798/* Requires ECN/ECT set on all packets */
799#define TCP_CONG_NEEDS_ECN 0x2
Stephen Hemminger164891a2007-04-23 22:26:16 -0700800
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700801struct tcp_congestion_ops {
802 struct list_head list;
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100803 u32 key;
804 u32 flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700805
806 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300807 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700808 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300809 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700810
811 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300812 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700813 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700814 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700815 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300816 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700817 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300818 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Florian Westphal7354c8c2014-09-26 22:37:34 +0200819 /* call when ack arrives (optional) */
820 void (*in_ack_event)(struct sock *sk, u32 flags);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700821 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300822 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700823 /* hook for packet ack accounting (optional) */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700824 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300825 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300826 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700827
828 char name[TCP_CA_NAME_MAX];
829 struct module *owner;
830};
831
Joe Perches5c9f3022013-09-23 11:33:32 -0700832int tcp_register_congestion_control(struct tcp_congestion_ops *type);
833void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700834
Florian Westphal55d86942014-09-26 22:37:32 +0200835void tcp_assign_congestion_control(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700836void tcp_init_congestion_control(struct sock *sk);
837void tcp_cleanup_congestion_control(struct sock *sk);
838int tcp_set_default_congestion_control(const char *name);
839void tcp_get_default_congestion_control(char *name);
840void tcp_get_available_congestion_control(char *buf, size_t len);
841void tcp_get_allowed_congestion_control(char *buf, size_t len);
842int tcp_set_allowed_congestion_control(char *allowed);
843int tcp_set_congestion_control(struct sock *sk, const char *name);
Neal Cardwelle73ebb082015-01-28 20:01:35 -0500844u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
845void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700846
Joe Perches5c9f3022013-09-23 11:33:32 -0700847u32 tcp_reno_ssthresh(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -0700848void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -0700849extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700850
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100851struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
852u32 tcp_ca_get_key_by_name(const char *name);
Daniel Borkmannea697632015-01-05 23:57:47 +0100853#ifdef CONFIG_INET
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100854char *tcp_ca_get_name_by_key(u32 key, char *buffer);
Daniel Borkmannea697632015-01-05 23:57:47 +0100855#else
856static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
857{
858 return NULL;
859}
860#endif
Daniel Borkmannc5c6a8a2015-01-05 23:57:46 +0100861
Daniel Borkmann30e502a2014-09-26 22:37:33 +0200862static inline bool tcp_ca_needs_ecn(const struct sock *sk)
863{
864 const struct inet_connection_sock *icsk = inet_csk(sk);
865
866 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
867}
868
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300869static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700870{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300871 struct inet_connection_sock *icsk = inet_csk(sk);
872
873 if (icsk->icsk_ca_ops->set_state)
874 icsk->icsk_ca_ops->set_state(sk, ca_state);
875 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700876}
877
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300878static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700879{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300880 const struct inet_connection_sock *icsk = inet_csk(sk);
881
882 if (icsk->icsk_ca_ops->cwnd_event)
883 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700884}
885
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300886/* These functions determine how the current flow behaves in respect of SACK
887 * handling. SACK is negotiated with the peer, and therefore it can vary
888 * between different flows.
889 *
890 * tcp_is_sack - SACK enabled
891 * tcp_is_reno - No SACK
892 * tcp_is_fack - FACK enabled, implies SACK enabled
893 */
894static inline int tcp_is_sack(const struct tcp_sock *tp)
895{
896 return tp->rx_opt.sack_ok;
897}
898
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000899static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300900{
901 return !tcp_is_sack(tp);
902}
903
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000904static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300905{
Vijay Subramanianab562222011-12-20 13:23:24 +0000906 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300907}
908
909static inline void tcp_enable_fack(struct tcp_sock *tp)
910{
Vijay Subramanianab562222011-12-20 13:23:24 +0000911 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300912}
913
Yuchung Chengeed530b2012-05-02 13:30:03 +0000914/* TCP early-retransmit (ER) is similar to but more conservative than
915 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
916 */
917static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
918{
919 tp->do_early_retrans = sysctl_tcp_early_retrans &&
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000920 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
921 sysctl_tcp_reordering == 3;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000922}
923
924static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
925{
926 tp->do_early_retrans = 0;
927}
928
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300929static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
930{
931 return tp->sacked_out + tp->lost_out;
932}
933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934/* This determines how many packets are "in the network" to the best
935 * of our knowledge. In many cases it is conservative, but where
936 * detailed information is available from the receiver (via SACK
937 * blocks etc.) we can make more aggressive calculations.
938 *
939 * Use this for decisions involving congestion control, use just
940 * tp->packets_out to determine if the send queue is empty or not.
941 *
942 * Read this equation as:
943 *
944 * "Packets sent once on transmission queue" MINUS
945 * "Packets left network, but not honestly ACKed yet" PLUS
946 * "Packets fast retransmitted"
947 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800948static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300950 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -0700953#define TCP_INFINITE_SSTHRESH 0x7fffffff
954
955static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
956{
957 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
958}
959
Yuchung Cheng684bad12012-09-02 17:38:04 +0000960static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
961{
962 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
963 (1 << inet_csk(sk)->icsk_ca_state);
964}
965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +0000967 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 * ssthresh.
969 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300970static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300972 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400973
Yuchung Cheng684bad12012-09-02 17:38:04 +0000974 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 return tp->snd_ssthresh;
976 else
977 return max(tp->snd_ssthresh,
978 ((tp->snd_cwnd >> 1) +
979 (tp->snd_cwnd >> 2)));
980}
981
Ilpo Järvinenb9c45952007-07-27 16:36:17 +0300982/* Use define here intentionally to get WARN_ON location shown at the caller */
983#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Christoph Paasch5ee2c942014-07-14 16:58:32 +0200985void tcp_enter_cwr(struct sock *sk);
Joe Perches5c9f3022013-09-23 11:33:32 -0700986__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Neal Cardwell6b5a5c02011-11-21 17:15:14 +0000988/* The maximum number of MSS of available cwnd for which TSO defers
989 * sending if not using sysctl_tcp_tso_win_divisor.
990 */
991static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
992{
993 return 3;
994}
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996/* Slow start with delack produces 3 packets of burst, so that
John Heffnerdd9e0dd2008-04-15 15:26:39 -0700997 * it is safe "de facto". This will be the default - same as
998 * the default reordering threshold - but if reordering increases,
999 * we must be able to allow cwnd to burst at least this much in order
1000 * to not pull it back when holes are filled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 */
1002static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
1003{
John Heffnerdd9e0dd2008-04-15 15:26:39 -07001004 return tp->reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
Ilpo Järvinen90840de2007-12-31 04:48:41 -08001007/* Returns end sequence number of the receiver's advertised window */
1008static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1009{
1010 return tp->snd_una + tp->snd_wnd;
1011}
Eric Dumazete114a712014-04-30 11:58:13 -07001012
1013/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1014 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -04001015 * it was fully used previously. And that's exactly what we do in
1016 * congestion avoidance mode. But in slow start we allow cwnd to grow
1017 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -07001018 * Example :
1019 * cwnd is 10 (IW10), but application sends 9 frames.
1020 * We allow cwnd to reach 18 when all frames are ACKed.
1021 * This check is safe because it's as aggressive as slow start which already
1022 * risks 100% overshoot. The advantage is that we discourage application to
1023 * either send more filler packets or data to artificially blow up the cwnd
1024 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -07001025 */
Eric Dumazet24901552014-05-02 21:18:05 -07001026static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -07001027{
1028 const struct tcp_sock *tp = tcp_sk(sk);
1029
Neal Cardwellca8a2262014-05-22 10:41:08 -04001030 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1031 if (tp->snd_cwnd <= tp->snd_ssthresh)
1032 return tp->snd_cwnd < 2 * tp->max_packets_out;
1033
1034 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -07001035}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -08001036
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001037static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038{
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001039 const struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001040 const struct inet_connection_sock *icsk = inet_csk(sk);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -07001041
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001042 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -07001043 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1044 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045}
1046
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001047static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
1049 tp->snd_wl1 = seq;
1050}
1051
Hantzis Fotisee7537b2009-03-02 22:42:02 -08001052static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
1054 tp->snd_wl1 = seq;
1055}
1056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057/*
1058 * Calculate(/check) TCP checksum
1059 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001060static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1061 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
1063 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1064}
1065
Al Virob51655b2006-11-14 21:40:42 -08001066static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Herbert Xufb286bb2005-11-10 13:01:24 -08001068 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069}
1070
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001071static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072{
Herbert Xu60476372007-04-09 11:59:39 -07001073 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 __tcp_checksum_complete(skb);
1075}
1076
1077/* Prequeue for VJ style copy to user, combined with checksumming. */
1078
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001079static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
1081 tp->ucopy.task = NULL;
1082 tp->ucopy.len = 0;
1083 tp->ucopy.memory = 0;
1084 skb_queue_head_init(&tp->ucopy.prequeue);
1085}
1086
Joe Perches5c9f3022013-09-23 11:33:32 -07001087bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089#undef STATE_TRACE
1090
1091#ifdef STATE_TRACE
1092static const char *statename[]={
1093 "Unused","Established","Syn Sent","Syn Recv",
1094 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1095 "Close Wait","Last ACK","Listen","Closing"
1096};
1097#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001098void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Joe Perches5c9f3022013-09-23 11:33:32 -07001100void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001102static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103{
1104 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 rx_opt->num_sacks = 0;
1106}
1107
Joe Perches5c9f3022013-09-23 11:33:32 -07001108u32 tcp_default_init_rwnd(u32 mss);
Yuchung Cheng85f16522013-06-11 15:35:32 -07001109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001111void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1112 __u32 *window_clamp, int wscale_ok,
1113 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115static inline int tcp_win_from_space(int space)
1116{
1117 return sysctl_tcp_adv_win_scale<=0 ?
1118 (space>>(-sysctl_tcp_adv_win_scale)) :
1119 space - (space>>sysctl_tcp_adv_win_scale);
1120}
1121
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001122/* Note: caller must be prepared to deal with negative returns */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123static inline int tcp_space(const struct sock *sk)
1124{
1125 return tcp_win_from_space(sk->sk_rcvbuf -
1126 atomic_read(&sk->sk_rmem_alloc));
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001127}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
1129static inline int tcp_full_space(const struct sock *sk)
1130{
Kenjiro Nakayama105970f2014-10-20 18:15:50 +09001131 return tcp_win_from_space(sk->sk_rcvbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132}
1133
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001134static inline void tcp_openreq_init(struct request_sock *req,
1135 struct tcp_options_received *rx_opt,
Octavian Purdilae0f802f2014-06-17 11:25:37 +03001136 struct sk_buff *skb, struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001138 struct inet_request_sock *ireq = inet_rsk(req);
1139
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Florian Westphal4dfc2812008-04-10 03:12:40 -07001141 req->cookie_ts = 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001142 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Jerry Chu10467162012-08-31 12:29:11 +00001143 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
Neal Cardwell86c6a2c2014-06-30 15:09:49 -04001144 tcp_rsk(req)->snt_synack = tcp_time_stamp;
Neal Cardwella9b2c062015-02-06 16:04:39 -05001145 tcp_rsk(req)->last_oow_ack_time = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 req->mss = rx_opt->mss_clamp;
1147 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001148 ireq->tstamp_ok = rx_opt->tstamp_ok;
1149 ireq->sack_ok = rx_opt->sack_ok;
1150 ireq->snd_wscale = rx_opt->snd_wscale;
1151 ireq->wscale_ok = rx_opt->wscale_ok;
1152 ireq->acked = 0;
1153 ireq->ecn_ok = 0;
Eric Dumazet634fb9792013-10-09 15:21:29 -07001154 ireq->ir_rmt_port = tcp_hdr(skb)->source;
Eric Dumazetb44084c2013-10-10 00:04:37 -07001155 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
Octavian Purdilae0f802f2014-06-17 11:25:37 +03001156 ireq->ir_mark = inet_request_mark(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157}
1158
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001159extern void tcp_openreq_init_rwin(struct request_sock *req,
1160 struct sock *sk, struct dst_entry *dst);
1161
Joe Perches5c9f3022013-09-23 11:33:32 -07001162void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1165{
1166 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1167}
1168
1169static inline int keepalive_time_when(const struct tcp_sock *tp)
1170{
1171 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1172}
1173
Eric Dumazetdf19a622009-08-28 23:48:54 -07001174static inline int keepalive_probes(const struct tcp_sock *tp)
1175{
1176 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1177}
1178
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001179static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1180{
1181 const struct inet_connection_sock *icsk = &tp->inet_conn;
1182
1183 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1184 tcp_time_stamp - tp->rcv_tstamp);
1185}
1186
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001187static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001189 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1190 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001192 if (fin_timeout < (rto << 2) - (rto >> 1))
1193 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195 return fin_timeout;
1196}
1197
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001198static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1199 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001201 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001202 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001203 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001204 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001205 /*
1206 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1207 * then following tcp messages have valid values. Ignore 0 value,
1208 * or else 'negative' tsval might forbid us to accept their packets.
1209 */
1210 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001211 return true;
1212 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001213}
1214
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001215static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1216 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001217{
1218 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001219 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221 /* RST segments are not recommended to carry timestamp,
1222 and, if they do, it is recommended to ignore PAWS because
1223 "their cleanup function should take precedence over timestamps."
1224 Certainly, it is mistake. It is necessary to understand the reasons
1225 of this constraint to relax it: if peer reboots, clock may go
1226 out-of-sync and half-open connections will not be reset.
1227 Actually, the problem would be not existing if all
1228 the implementations followed draft about maintaining clock
1229 via reboots. Linux-2.2 DOES NOT!
1230
1231 However, we can relax time bounds for RST segments to MSL.
1232 */
James Morris9d729f72007-03-04 16:12:44 -08001233 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001234 return false;
1235 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236}
1237
Neal Cardwell032ee422015-02-06 16:04:38 -05001238/* Return true if we're currently rate-limiting out-of-window ACKs and
1239 * thus shouldn't send a dupack right now. We rate-limit dupacks in
1240 * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
1241 * attacks that send repeated SYNs or ACKs for the same connection. To
1242 * do this, we do not send a duplicate SYNACK or ACK if the remote
1243 * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
1244 */
1245static inline bool tcp_oow_rate_limited(struct net *net,
1246 const struct sk_buff *skb,
1247 int mib_idx, u32 *last_oow_ack_time)
1248{
1249 /* Data packets without SYNs are not likely part of an ACK loop. */
1250 if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) &&
1251 !tcp_hdr(skb)->syn)
1252 goto not_rate_limited;
1253
1254 if (*last_oow_ack_time) {
1255 s32 elapsed = (s32)(tcp_time_stamp - *last_oow_ack_time);
1256
1257 if (0 <= elapsed && elapsed < sysctl_tcp_invalid_ratelimit) {
1258 NET_INC_STATS_BH(net, mib_idx);
1259 return true; /* rate-limited: don't send yet! */
1260 }
1261 }
1262
1263 *last_oow_ack_time = tcp_time_stamp;
1264
1265not_rate_limited:
1266 return false; /* not rate-limited: go ahead, send dupack now! */
1267}
1268
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001269static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
1271 /* See RFC 2012 */
Pavel Emelyanovcf1100a2008-07-16 20:27:38 -07001272 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1273 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1274 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1275 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276}
1277
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001278/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001279static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001280{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001281 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001282}
1283
1284static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1285{
1286 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001287 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001288}
1289
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001290/* MD5 Signature */
1291struct crypto_hash;
1292
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001293union tcp_md5_addr {
1294 struct in_addr a4;
1295#if IS_ENABLED(CONFIG_IPV6)
1296 struct in6_addr a6;
1297#endif
1298};
1299
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001300/* - key database */
1301struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001302 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001303 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001304 u8 family; /* AF_INET or AF_INET6 */
1305 union tcp_md5_addr addr;
1306 u8 key[TCP_MD5SIG_MAXKEYLEN];
1307 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001308};
1309
1310/* - sock block */
1311struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001312 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001313 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001314};
1315
1316/* - pseudo header */
1317struct tcp4_pseudohdr {
1318 __be32 saddr;
1319 __be32 daddr;
1320 __u8 pad;
1321 __u8 protocol;
1322 __be16 len;
1323};
1324
1325struct tcp6_pseudohdr {
1326 struct in6_addr saddr;
1327 struct in6_addr daddr;
1328 __be32 len;
1329 __be32 protocol; /* including padding */
1330};
1331
1332union tcp_md5sum_block {
1333 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001334#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001335 struct tcp6_pseudohdr ip6;
1336#endif
1337};
1338
1339/* - pool: digest algorithm, hash description and scratch buffer */
1340struct tcp_md5sig_pool {
1341 struct hash_desc md5_desc;
1342 union tcp_md5sum_block md5_blk;
1343};
1344
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001345/* - functions */
Joe Perches5c9f3022013-09-23 11:33:32 -07001346int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1347 const struct sock *sk, const struct request_sock *req,
1348 const struct sk_buff *skb);
1349int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1350 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1351int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1352 int family);
1353struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001354 struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001355
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001356#ifdef CONFIG_TCP_MD5SIG
Joe Perches5c9f3022013-09-23 11:33:32 -07001357struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1358 const union tcp_md5_addr *addr,
1359 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001360#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001361#else
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001362static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1363 const union tcp_md5_addr *addr,
1364 int family)
1365{
1366 return NULL;
1367}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001368#define tcp_twsk_md5_key(twsk) NULL
1369#endif
1370
Joe Perches5c9f3022013-09-23 11:33:32 -07001371bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001372
Joe Perches5c9f3022013-09-23 11:33:32 -07001373struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001374static inline void tcp_put_md5sig_pool(void)
1375{
1376 local_bh_enable();
1377}
Eric Dumazet35790c02010-05-16 00:34:04 -07001378
Joe Perches5c9f3022013-09-23 11:33:32 -07001379int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1380int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1381 unsigned int header_len);
1382int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1383 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001384
Jerry Chu10467162012-08-31 12:29:11 +00001385/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001386void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1387 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1388 unsigned long *last_syn_loss);
1389void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1390 struct tcp_fastopen_cookie *cookie, bool syn_lost);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001391struct tcp_fastopen_request {
1392 /* Fast Open cookie. Size 0 means a cookie request */
1393 struct tcp_fastopen_cookie cookie;
1394 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001395 size_t size;
1396 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001397};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001398void tcp_free_fastopen_req(struct tcp_sock *tp);
1399
Jerry Chu10467162012-08-31 12:29:11 +00001400extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1401int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001402bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1403 struct request_sock *req,
1404 struct tcp_fastopen_cookie *foc,
1405 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001406void tcp_fastopen_init_key_once(bool publish);
Jerry Chu10467162012-08-31 12:29:11 +00001407#define TCP_FASTOPEN_KEY_LENGTH 16
1408
1409/* Fastopen key context */
1410struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001411 struct crypto_cipher *tfm;
1412 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1413 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001414};
1415
David S. Millerfe067e82007-03-07 12:12:44 -08001416/* write queue abstraction */
1417static inline void tcp_write_queue_purge(struct sock *sk)
1418{
1419 struct sk_buff *skb;
1420
1421 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001422 sk_wmem_free_skb(sk, skb);
1423 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001424 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001425}
1426
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001427static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001428{
David S. Millercd07a8e2008-09-23 00:50:13 -07001429 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001430}
1431
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001432static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001433{
David S. Millercd07a8e2008-09-23 00:50:13 -07001434 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001435}
1436
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001437static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1438 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001439{
David S. Millercd07a8e2008-09-23 00:50:13 -07001440 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001441}
1442
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001443static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1444 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001445{
1446 return skb_queue_prev(&sk->sk_write_queue, skb);
1447}
1448
David S. Millerfe067e82007-03-07 12:12:44 -08001449#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001450 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001451
1452#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001453 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001454
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001455#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001456 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001457
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001458static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001459{
1460 return sk->sk_send_head;
1461}
1462
David S. Millercd07a8e2008-09-23 00:50:13 -07001463static inline bool tcp_skb_is_last(const struct sock *sk,
1464 const struct sk_buff *skb)
1465{
1466 return skb_queue_is_last(&sk->sk_write_queue, skb);
1467}
1468
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001469static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001470{
David S. Millercd07a8e2008-09-23 00:50:13 -07001471 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001472 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001473 else
1474 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001475}
1476
1477static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1478{
1479 if (sk->sk_send_head == skb_unlinked)
1480 sk->sk_send_head = NULL;
1481}
1482
1483static inline void tcp_init_send_head(struct sock *sk)
1484{
1485 sk->sk_send_head = NULL;
1486}
1487
1488static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1489{
1490 __skb_queue_tail(&sk->sk_write_queue, skb);
1491}
1492
1493static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1494{
1495 __tcp_add_write_queue_tail(sk, skb);
1496
1497 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001498 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001499 sk->sk_send_head = skb;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001500
1501 if (tcp_sk(sk)->highest_sack == NULL)
1502 tcp_sk(sk)->highest_sack = skb;
1503 }
David S. Millerfe067e82007-03-07 12:12:44 -08001504}
1505
1506static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1507{
1508 __skb_queue_head(&sk->sk_write_queue, skb);
1509}
1510
1511/* Insert buff after skb on the write queue of sk. */
1512static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1513 struct sk_buff *buff,
1514 struct sock *sk)
1515{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001516 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001517}
1518
David S. Miller43f59c82008-09-21 21:28:51 -07001519/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001520static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1521 struct sk_buff *skb,
1522 struct sock *sk)
1523{
David S. Miller43f59c82008-09-21 21:28:51 -07001524 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001525
1526 if (sk->sk_send_head == skb)
1527 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001528}
1529
1530static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1531{
1532 __skb_unlink(skb, &sk->sk_write_queue);
1533}
1534
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001535static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001536{
1537 return skb_queue_empty(&sk->sk_write_queue);
1538}
1539
Krishna Kumar12d50c42009-12-08 22:26:13 +00001540static inline void tcp_push_pending_frames(struct sock *sk)
1541{
1542 if (tcp_send_head(sk)) {
1543 struct tcp_sock *tp = tcp_sk(sk);
1544
1545 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1546 }
1547}
1548
Neal Cardwellecb97192012-02-27 17:52:52 -05001549/* Start sequence of the skb just after the highest skb with SACKed
1550 * bit, valid only if sacked_out > 0 or when the caller has ensured
1551 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001552 */
1553static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1554{
1555 if (!tp->sacked_out)
1556 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001557
1558 if (tp->highest_sack == NULL)
1559 return tp->snd_nxt;
1560
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001561 return TCP_SKB_CB(tp->highest_sack)->seq;
1562}
1563
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001564static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1565{
1566 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1567 tcp_write_queue_next(sk, skb);
1568}
1569
1570static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1571{
1572 return tcp_sk(sk)->highest_sack;
1573}
1574
1575static inline void tcp_highest_sack_reset(struct sock *sk)
1576{
1577 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1578}
1579
1580/* Called when old skb is about to be deleted (to be combined with new skb) */
1581static inline void tcp_highest_sack_combine(struct sock *sk,
1582 struct sk_buff *old,
1583 struct sk_buff *new)
1584{
1585 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1586 tcp_sk(sk)->highest_sack = new;
1587}
1588
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001589/* Determines whether this is a thin stream (which may suffer from
1590 * increased latency). Used to trigger latency-reducing mechanisms.
1591 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001592static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001593{
1594 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1595}
1596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597/* /proc */
1598enum tcp_seq_states {
1599 TCP_SEQ_STATE_LISTENING,
1600 TCP_SEQ_STATE_OPENREQ,
1601 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602};
1603
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001604int tcp_seq_open(struct inode *inode, struct file *file);
1605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001607 char *name;
1608 sa_family_t family;
1609 const struct file_operations *seq_fops;
1610 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611};
1612
1613struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001614 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 sa_family_t family;
1616 enum tcp_seq_states state;
1617 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001618 int bucket, offset, sbucket, num;
1619 kuid_t uid;
Tom Herberta8b690f2010-06-07 00:43:42 -07001620 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621};
1622
Joe Perches5c9f3022013-09-23 11:33:32 -07001623int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1624void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001626extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001627extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001628
Joe Perches5c9f3022013-09-23 11:33:32 -07001629void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001630
Eric Dumazet28be6e02013-10-18 10:36:17 -07001631struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001632 netdev_features_t features);
1633struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1634int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001635
Joe Perches5c9f3022013-09-23 11:33:32 -07001636void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001637
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001638static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1639{
1640 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1641}
1642
1643static inline bool tcp_stream_memory_free(const struct sock *sk)
1644{
1645 const struct tcp_sock *tp = tcp_sk(sk);
1646 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1647
1648 return notsent_bytes < tcp_notsent_lowat(tp);
1649}
1650
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001651#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001652int tcp4_proc_init(void);
1653void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001654#endif
1655
Octavian Purdila5db92c92014-06-25 17:09:59 +03001656int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
Octavian Purdila1fb6f152014-06-25 17:10:02 +03001657int tcp_conn_request(struct request_sock_ops *rsk_ops,
1658 const struct tcp_request_sock_ops *af_ops,
1659 struct sock *sk, struct sk_buff *skb);
Octavian Purdila5db92c92014-06-25 17:09:59 +03001660
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001661/* TCP af-specific functions */
1662struct tcp_sock_af_ops {
1663#ifdef CONFIG_TCP_MD5SIG
1664 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1665 struct sock *addr_sk);
1666 int (*calc_md5_hash) (char *location,
1667 struct tcp_md5sig_key *md5,
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001668 const struct sock *sk,
1669 const struct request_sock *req,
1670 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001671 int (*md5_parse) (struct sock *sk,
1672 char __user *optval,
1673 int optlen);
1674#endif
1675};
1676
1677struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001678 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001679#ifdef CONFIG_TCP_MD5SIG
1680 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1681 struct request_sock *req);
John Dykstrae3afe7b2009-07-16 05:04:51 +00001682 int (*calc_md5_hash) (char *location,
1683 struct tcp_md5sig_key *md5,
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001684 const struct sock *sk,
1685 const struct request_sock *req,
1686 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001687#endif
Octavian Purdila16bea702014-06-25 17:09:53 +03001688 void (*init_req)(struct request_sock *req, struct sock *sk,
1689 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001690#ifdef CONFIG_SYN_COOKIES
1691 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1692 __u16 *mss);
1693#endif
Octavian Purdilad94e0412014-06-25 17:09:55 +03001694 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1695 const struct request_sock *req,
1696 bool *strict);
Octavian Purdila936b8bd2014-06-25 17:09:57 +03001697 __u32 (*init_seq)(const struct sk_buff *skb);
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001698 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1699 struct flowi *fl, struct request_sock *req,
1700 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
Octavian Purdila695da142014-06-25 17:10:01 +03001701 void (*queue_hash_add)(struct sock *sk, struct request_sock *req,
1702 const unsigned long timeout);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001703};
1704
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001705#ifdef CONFIG_SYN_COOKIES
1706static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1707 struct sock *sk, struct sk_buff *skb,
1708 __u16 *mss)
1709{
1710 return ops->cookie_init_seq(sk, skb, mss);
1711}
1712#else
1713static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1714 struct sock *sk, struct sk_buff *skb,
1715 __u16 *mss)
1716{
1717 return 0;
1718}
1719#endif
1720
Joe Perches5c9f3022013-09-23 11:33:32 -07001721int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001722
Joe Perches5c9f3022013-09-23 11:33:32 -07001723void tcp_v4_init(void);
1724void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001725
Cong Wange25f8662014-10-15 14:33:21 -07001726/*
1727 * Save and compile IPv4 options, return a pointer to it
1728 */
1729static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1730{
1731 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1732 struct ip_options_rcu *dopt = NULL;
1733
Cong Wang461b74c2014-10-15 14:33:22 -07001734 if (opt->optlen) {
Cong Wange25f8662014-10-15 14:33:21 -07001735 int opt_size = sizeof(*dopt) + opt->optlen;
1736
1737 dopt = kmalloc(opt_size, GFP_ATOMIC);
1738 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1739 kfree(dopt);
1740 dopt = NULL;
1741 }
1742 }
1743 return dopt;
1744}
1745
Eric Dumazet98781962015-02-03 18:31:53 -08001746/* locally generated TCP pure ACKs have skb->truesize == 2
1747 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1748 * This is much faster than dissecting the packet to find out.
1749 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1750 */
1751static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1752{
1753 return skb->truesize == 2;
1754}
1755
1756static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1757{
1758 skb->truesize = 2;
1759}
1760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761#endif /* _TCP_H */