blob: 30fe98bc8957e5b689e413fd95ccbee836f17a7d [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>
Chris Leech97fc2f02006-05-23 17:55:33 -070030#include <linux/dmaengine.h>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080031#include <linux/crypto.h>
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080032#include <linux/cryptohash.h>
William Allen Simpson435cf552009-12-02 18:17:05 +000033#include <linux/kref.h>
Eric Dumazet740b0f12014-02-26 14:02:48 -080034#include <linux/ktime.h>
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070035
36#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -070037#include <net/inet_timewait_sock.h>
Arnaldo Carvalho de Melo77d8bf92005-08-09 20:00:51 -070038#include <net/inet_hashtables.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <net/checksum.h>
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -070040#include <net/request_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <net/sock.h>
42#include <net/snmp.h>
43#include <net/ip.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070044#include <net/tcp_states.h>
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -070045#include <net/inet_ecn.h>
Satoru SATOH0c266892009-05-04 11:11:01 -070046#include <net/dst.h>
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070047
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/seq_file.h>
Glauber Costa180d8cd2011-12-11 21:47:02 +000049#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Arnaldo Carvalho de Melo6e04e022005-08-09 20:07:35 -070051extern struct inet_hashinfo tcp_hashinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Eric Dumazetdd24c002008-11-25 21:17:14 -080053extern struct percpu_counter tcp_orphan_count;
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#define MAX_TCP_HEADER (128 + MAX_HEADER)
Adam Langley33ad7982008-07-19 00:04:31 -070057#define MAX_TCP_OPTION_SPACE 40
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59/*
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
62 */
63#define MAX_TCP_WINDOW 32767U
64
65/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66#define TCP_MIN_MSS 88U
67
John Heffner5d424d52006-03-20 17:53:41 -080068/* The least MTU to use for probing */
69#define TCP_BASE_MSS 512
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071/* After receiving this amount of duplicate ACKs fast retransmit starts. */
72#define TCP_FASTRETRANS_THRESH 3
73
74/* Maximal reordering. */
75#define TCP_MAX_REORDERING 127
76
77/* Maximal number of ACKs sent quickly to accelerate slow-start. */
78#define TCP_MAX_QUICKACKS 16U
79
80/* urg_data states */
81#define TCP_URG_VALID 0x0100
82#define TCP_URG_NOTYET 0x0200
83#define TCP_URG_READ 0x0400
84
85#define TCP_RETR1 3 /*
86 * This is how many retries it does before it
87 * tries to figure out if the gateway is
88 * down. Minimal RFC value is 3; it corresponds
89 * to ~3sec-8min depending on RTO.
90 */
91
92#define TCP_RETR2 15 /*
93 * This should take at least
94 * 90 minutes to time out.
95 * RFC1122 says that the limit is 100 sec.
96 * 15 is ~13-30min depending on RTO.
97 */
98
Alex Bergmann6c9ff972012-08-31 02:48:31 +000099#define TCP_SYN_RETRIES 6 /* This is how many retries are done
100 * when active opening a connection.
101 * RFC1122 says the minimum retry MUST
102 * be at least 180secs. Nevertheless
103 * this value is corresponding to
104 * 63secs of retransmission with the
105 * current initial RTO.
106 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Alex Bergmann6c9ff972012-08-31 02:48:31 +0000108#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
109 * when passive opening a connection.
110 * This is corresponding to 31secs of
111 * retransmission with the current
112 * initial RTO.
113 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
116 * state, about 60 seconds */
117#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
118 /* BSD style FIN_WAIT2 deadlock breaker.
119 * It used to be 3min, new value is 60sec,
120 * to combine FIN-WAIT-2 timeout with
121 * TIME-WAIT timer.
122 */
123
124#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
125#if HZ >= 100
126#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
127#define TCP_ATO_MIN ((unsigned)(HZ/25))
128#else
129#define TCP_DELACK_MIN 4U
130#define TCP_ATO_MIN 4U
131#endif
132#define TCP_RTO_MAX ((unsigned)(120*HZ))
133#define TCP_RTO_MIN ((unsigned)(HZ/5))
Eric Dumazetfd4f2ce2012-04-12 19:48:40 +0000134#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
Jerry Chu9ad7c042011-06-08 11:08:38 +0000135#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
136 * used as a fallback RTO for the
137 * initial data transmission if no
138 * valid RTT sample has been acquired,
139 * most likely due to retrans in 3WHS.
140 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
143 * for local resources.
144 */
145
146#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
147#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
148#define TCP_KEEPALIVE_INTVL (75*HZ)
149
150#define MAX_TCP_KEEPIDLE 32767
151#define MAX_TCP_KEEPINTVL 32767
152#define MAX_TCP_KEEPCNT 127
153#define MAX_TCP_SYNCNT 127
154
155#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
158#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
159 * after this time. It should be equal
160 * (or greater than) TCP_TIMEWAIT_LEN
161 * to provide reliability equal to one
162 * provided by timewait state.
163 */
164#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
165 * timestamps. It must be less than
166 * minimal timewait lifetime.
167 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/*
169 * TCP option
170 */
171
172#define TCPOPT_NOP 1 /* Padding */
173#define TCPOPT_EOL 0 /* End of options */
174#define TCPOPT_MSS 2 /* Segment size negotiating */
175#define TCPOPT_WINDOW 3 /* Window scaling */
176#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
177#define TCPOPT_SACK 5 /* SACK Block */
178#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800179#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000180#define TCPOPT_EXP 254 /* Experimental */
181/* Magic number to be after the option value for sharing TCP
182 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
183 */
184#define TCPOPT_FASTOPEN_MAGIC 0xF989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
186/*
187 * TCP option lengths
188 */
189
190#define TCPOLEN_MSS 4
191#define TCPOLEN_WINDOW 3
192#define TCPOLEN_SACK_PERM 2
193#define TCPOLEN_TIMESTAMP 10
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800194#define TCPOLEN_MD5SIG 18
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000195#define TCPOLEN_EXP_FASTOPEN_BASE 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197/* But this is what stacks really send out. */
198#define TCPOLEN_TSTAMP_ALIGNED 12
199#define TCPOLEN_WSCALE_ALIGNED 4
200#define TCPOLEN_SACKPERM_ALIGNED 4
201#define TCPOLEN_SACK_BASE 2
202#define TCPOLEN_SACK_BASE_ALIGNED 4
203#define TCPOLEN_SACK_PERBLOCK 8
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800204#define TCPOLEN_MD5SIG_ALIGNED 20
Adam Langley33ad7982008-07-19 00:04:31 -0700205#define TCPOLEN_MSS_ALIGNED 4
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207/* Flags in tp->nonagle */
208#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
209#define TCP_NAGLE_CORK 2 /* Socket is corked */
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800210#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000212/* TCP thin-stream limits */
213#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
214
David S. Miller7eb38522011-02-05 18:13:45 -0800215/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
David S. Miller442b9632011-02-02 17:05:11 -0800216#define TCP_INIT_CWND 10
217
Yuchung Chengcf60af02012-07-19 06:43:09 +0000218/* Bit Flags for sysctl_tcp_fastopen */
219#define TFO_CLIENT_ENABLE 1
Jerry Chu10467162012-08-31 12:29:11 +0000220#define TFO_SERVER_ENABLE 2
Yuchung Cheng67da22d2012-07-19 06:43:11 +0000221#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
Yuchung Chengcf60af02012-07-19 06:43:09 +0000222
Jerry Chu10467162012-08-31 12:29:11 +0000223/* Accept SYN data w/o any cookie option */
224#define TFO_SERVER_COOKIE_NOT_REQD 0x200
225
226/* Force enable TFO on all listeners, i.e., not requiring the
227 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
228 */
229#define TFO_SERVER_WO_SOCKOPT1 0x400
230#define TFO_SERVER_WO_SOCKOPT2 0x800
Jerry Chu10467162012-08-31 12:29:11 +0000231
Arnaldo Carvalho de Melo295ff7e2005-08-09 20:44:40 -0700232extern struct inet_timewait_death_row tcp_death_row;
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/* sysctl variables for tcp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235extern int sysctl_tcp_timestamps;
236extern int sysctl_tcp_window_scaling;
237extern int sysctl_tcp_sack;
238extern int sysctl_tcp_fin_timeout;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239extern int sysctl_tcp_keepalive_time;
240extern int sysctl_tcp_keepalive_probes;
241extern int sysctl_tcp_keepalive_intvl;
242extern int sysctl_tcp_syn_retries;
243extern int sysctl_tcp_synack_retries;
244extern int sysctl_tcp_retries1;
245extern int sysctl_tcp_retries2;
246extern int sysctl_tcp_orphan_retries;
247extern int sysctl_tcp_syncookies;
Yuchung Cheng2100c8d2012-07-19 06:43:05 +0000248extern int sysctl_tcp_fastopen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249extern int sysctl_tcp_retrans_collapse;
250extern int sysctl_tcp_stdurg;
251extern int sysctl_tcp_rfc1337;
252extern int sysctl_tcp_abort_on_overflow;
253extern int sysctl_tcp_max_orphans;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254extern int sysctl_tcp_fack;
255extern int sysctl_tcp_reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256extern int sysctl_tcp_dsack;
Eric W. Biedermana4fe34b2013-10-19 16:25:36 -0700257extern long sysctl_tcp_mem[3];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258extern int sysctl_tcp_wmem[3];
259extern int sysctl_tcp_rmem[3];
260extern int sysctl_tcp_app_win;
261extern int sysctl_tcp_adv_win_scale;
262extern int sysctl_tcp_tw_reuse;
263extern int sysctl_tcp_frto;
264extern int sysctl_tcp_low_latency;
Chris Leech95937822006-05-23 18:02:55 -0700265extern int sysctl_tcp_dma_copybreak;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266extern int sysctl_tcp_nometrics_save;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267extern int sysctl_tcp_moderate_rcvbuf;
268extern int sysctl_tcp_tso_win_divisor;
John Heffner5d424d52006-03-20 17:53:41 -0800269extern int sysctl_tcp_mtu_probing;
270extern int sysctl_tcp_base_mss;
Rick Jones15d99e02006-03-20 22:40:29 -0800271extern int sysctl_tcp_workaround_signed_windows;
David S. Miller35089bb2006-06-13 22:33:04 -0700272extern int sysctl_tcp_slow_start_after_idle;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000273extern int sysctl_tcp_thin_linear_timeouts;
Andreas Petlund7e380172010-02-18 04:48:19 +0000274extern int sysctl_tcp_thin_dupack;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000275extern int sysctl_tcp_early_retrans;
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000276extern int sysctl_tcp_limit_output_bytes;
Eric Dumazet282f23c2012-07-17 10:13:05 +0200277extern int sysctl_tcp_challenge_ack_limit;
Eric Dumazetc9bee3b72013-07-22 20:27:07 -0700278extern unsigned int sysctl_tcp_notsent_lowat;
Eric Dumazet95bd09e2013-08-27 05:46:32 -0700279extern int sysctl_tcp_min_tso_segs;
Eric Dumazetf54b3112013-12-05 22:36:05 -0800280extern int sysctl_tcp_autocorking;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
Eric Dumazet8d987e52010-11-09 23:24:26 +0000282extern atomic_long_t tcp_memory_allocated;
Eric Dumazet17483762008-11-25 21:16:35 -0800283extern struct percpu_counter tcp_sockets_allocated;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284extern int tcp_memory_pressure;
285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 * The next routines deal with comparing 32 bit unsigned ints
288 * and worry about wraparound (automatic with unsigned arithmetic).
289 */
290
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000291static inline bool before(__u32 seq1, __u32 seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
Gerrit Renker0d630cc2007-01-04 12:25:16 -0800293 return (__s32)(seq1-seq2) < 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
Gerrit Renker9a036b9c2006-12-20 10:25:55 -0800295#define after(seq2, seq1) before(seq1, seq2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
297/* is s2<=s1<=s3 ? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000298static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
300 return seq3 - seq2 >= seq1 - seq2;
301}
302
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800303static inline bool tcp_out_of_memory(struct sock *sk)
304{
305 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
306 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
307 return true;
308 return false;
309}
310
David S. Millerad1af0f2010-08-25 02:27:49 -0700311static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700312{
David S. Millerad1af0f2010-08-25 02:27:49 -0700313 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
314 int orphans = percpu_counter_read_positive(ocp);
315
316 if (orphans << shift > sysctl_tcp_max_orphans) {
317 orphans = percpu_counter_sum_positive(ocp);
318 if (orphans << shift > sysctl_tcp_max_orphans)
319 return true;
320 }
David S. Millerad1af0f2010-08-25 02:27:49 -0700321 return false;
Pavel Emelianove4fd5da2007-05-29 13:19:18 -0700322}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Joe Perches5c9f3022013-09-23 11:33:32 -0700324bool tcp_check_oom(struct sock *sk, int shift);
Arun Sharmaefcdbf22012-01-30 14:16:06 -0800325
Florian Westphala0f82f62009-04-19 09:43:48 +0000326/* syncookies: remember time of last synqueue overflow */
327static inline void tcp_synq_overflow(struct sock *sk)
328{
329 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
330}
331
332/* syncookies: no recent synqueue overflow on this listening socket? */
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000333static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
Florian Westphala0f82f62009-04-19 09:43:48 +0000334{
335 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
Jerry Chu9ad7c042011-06-08 11:08:38 +0000336 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
Florian Westphala0f82f62009-04-19 09:43:48 +0000337}
338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339extern struct proto tcp_prot;
340
Pavel Emelyanov57ef42d2008-07-18 04:02:08 -0700341#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
342#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
343#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
344#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 +0000345#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Joe Perches5c9f3022013-09-23 11:33:32 -0700347void tcp_tasklet_init(void);
Eric Dumazet46d3cea2012-07-11 05:50:31 +0000348
Joe Perches5c9f3022013-09-23 11:33:32 -0700349void tcp_v4_err(struct sk_buff *skb, u32);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
Joe Perches5c9f3022013-09-23 11:33:32 -0700351void tcp_shutdown(struct sock *sk, int how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
Joe Perches5c9f3022013-09-23 11:33:32 -0700353void tcp_v4_early_demux(struct sk_buff *skb);
354int tcp_v4_rcv(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
Joe Perches5c9f3022013-09-23 11:33:32 -0700356int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
357int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
358 size_t size);
359int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
360 int flags);
361void tcp_release_cb(struct sock *sk);
362void tcp_wfree(struct sk_buff *skb);
363void tcp_write_timer_handler(struct sock *sk);
364void tcp_delack_timer_handler(struct sock *sk);
365int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
366int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
367 const struct tcphdr *th, unsigned int len);
368void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
369 const struct tcphdr *th, unsigned int len);
370void tcp_rcv_space_adjust(struct sock *sk);
371void tcp_cleanup_rbuf(struct sock *sk, int copied);
372int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
373void tcp_twsk_destructor(struct sock *sk);
374ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
375 struct pipe_inode_info *pipe, size_t len,
376 unsigned int flags);
Jens Axboe9c55e012007-11-06 23:30:13 -0800377
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700378static inline void tcp_dec_quickack_mode(struct sock *sk,
379 const unsigned int pkts)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700381 struct inet_connection_sock *icsk = inet_csk(sk);
David S. Millerfc6415bc2005-07-05 15:17:45 -0700382
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700383 if (icsk->icsk_ack.quick) {
384 if (pkts >= icsk->icsk_ack.quick) {
385 icsk->icsk_ack.quick = 0;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700386 /* Leaving quickack mode we deflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700387 icsk->icsk_ack.ato = TCP_ATO_MIN;
David S. Millerfc6415bc2005-07-05 15:17:45 -0700388 } else
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700389 icsk->icsk_ack.quick -= pkts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 }
391}
392
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700393#define TCP_ECN_OK 1
394#define TCP_ECN_QUEUE_CWR 2
395#define TCP_ECN_DEMAND_CWR 4
Eric Dumazet7a269ff2011-09-22 20:02:19 +0000396#define TCP_ECN_SEEN 8
Ilpo Järvinenbdf1ee52007-05-27 02:04:16 -0700397
Eric Dumazetfd2c3ef2009-11-03 03:26:03 +0000398enum tcp_tw_status {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 TCP_TW_SUCCESS = 0,
400 TCP_TW_RST = 1,
401 TCP_TW_ACK = 2,
402 TCP_TW_SYN = 3
403};
404
405
Joe Perches5c9f3022013-09-23 11:33:32 -0700406enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
407 struct sk_buff *skb,
408 const struct tcphdr *th);
409struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
410 struct request_sock *req, struct request_sock **prev,
411 bool fastopen);
412int tcp_child_process(struct sock *parent, struct sock *child,
413 struct sk_buff *skb);
414void tcp_enter_loss(struct sock *sk, int how);
415void tcp_clear_retrans(struct tcp_sock *tp);
416void tcp_update_metrics(struct sock *sk);
417void tcp_init_metrics(struct sock *sk);
418void tcp_metrics_init(void);
419bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
420 bool paws_check);
421bool tcp_remember_stamp(struct sock *sk);
422bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
423void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
424void tcp_disable_fack(struct tcp_sock *tp);
425void tcp_close(struct sock *sk, long timeout);
426void tcp_init_sock(struct sock *sk);
427unsigned int tcp_poll(struct file *file, struct socket *sock,
428 struct poll_table_struct *wait);
429int tcp_getsockopt(struct sock *sk, int level, int optname,
430 char __user *optval, int __user *optlen);
431int tcp_setsockopt(struct sock *sk, int level, int optname,
432 char __user *optval, unsigned int optlen);
433int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000434 char __user *optval, int __user *optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700435int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
Changli Gao53d31762010-07-10 20:41:06 +0000436 char __user *optval, unsigned int optlen);
Joe Perches5c9f3022013-09-23 11:33:32 -0700437void tcp_set_keepalive(struct sock *sk, int val);
438void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
439int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
440 size_t len, int nonblock, int flags, int *addr_len);
441void tcp_parse_options(const struct sk_buff *skb,
442 struct tcp_options_received *opt_rx,
443 int estab, struct tcp_fastopen_cookie *foc);
444const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
YOSHIFUJI Hideaki7d5d5522008-04-17 12:29:53 +0900445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446/*
447 * TCP v4 functions exported for the inet6 API
448 */
449
Joe Perches5c9f3022013-09-23 11:33:32 -0700450void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
451int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
452struct sock *tcp_create_openreq_child(struct sock *sk,
453 struct request_sock *req,
454 struct sk_buff *skb);
455struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
456 struct request_sock *req,
457 struct dst_entry *dst);
458int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
459int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
460int tcp_connect(struct sock *sk);
461struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
462 struct request_sock *req,
463 struct tcp_fastopen_cookie *foc);
464int tcp_disconnect(struct sock *sk, int flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Pavel Emelyanov370816a2012-04-19 03:40:01 +0000466void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
Pavel Emelyanov292e8d82012-05-10 01:49:41 +0000467int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
Eric Dumazet63d02d12012-08-09 14:11:00 +0000468void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470/* From syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700471int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
472 u32 cookie);
473struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
474 struct ip_options *opt);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400475#ifdef CONFIG_SYN_COOKIES
Florian Westphal8c27bd72013-09-20 22:32:55 +0200476
Eric Dumazet63262312014-03-19 21:02:21 -0700477/* Syncookies use a monotonic timer which increments every 60 seconds.
Florian Westphal8c27bd72013-09-20 22:32:55 +0200478 * This counter is used both as a hash input and partially encoded into
479 * the cookie value. A cookie is only validated further if the delta
480 * between the current counter value and the encoded one is less than this,
Eric Dumazet63262312014-03-19 21:02:21 -0700481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
Florian Westphal8c27bd72013-09-20 22:32:55 +0200482 * the counter advances immediately after a cookie is generated).
483 */
484#define MAX_SYNCOOKIE_AGE 2
485
486static inline u32 tcp_cookie_time(void)
487{
Eric Dumazet63262312014-03-19 21:02:21 -0700488 u64 val = get_jiffies_64();
489
490 do_div(val, 60 * HZ);
491 return val;
Florian Westphal8c27bd72013-09-20 22:32:55 +0200492}
493
Joe Perches5c9f3022013-09-23 11:33:32 -0700494u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
495 u16 *mssp);
Octavian Purdila57b47552014-06-25 17:09:50 +0300496__u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
497 __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400498#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Joe Perches5c9f3022013-09-23 11:33:32 -0700500__u32 cookie_init_timestamp(struct request_sock *req);
501bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net,
502 bool *ecn_ok);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700503
Glenn Griffinc6aefaf2008-02-07 21:49:26 -0800504/* From net/ipv6/syncookies.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700505int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
506 u32 cookie);
507struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400508#ifdef CONFIG_SYN_COOKIES
Joe Perches5c9f3022013-09-23 11:33:32 -0700509u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
510 const struct tcphdr *th, u16 *mssp);
511__u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
512 __u16 *mss);
Eric Dumazete05c82d2011-09-18 21:02:55 -0400513#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514/* tcp_output.c */
515
Joe Perches5c9f3022013-09-23 11:33:32 -0700516void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
517 int nonagle);
518bool tcp_may_send_now(struct sock *sk);
519int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
520int tcp_retransmit_skb(struct sock *, struct sk_buff *);
521void tcp_retransmit_timer(struct sock *sk);
522void tcp_xmit_retransmit_queue(struct sock *);
523void tcp_simple_retransmit(struct sock *);
524int tcp_trim_head(struct sock *, struct sk_buff *, u32);
Octavian Purdila6cc55e02014-06-06 17:32:37 +0300525int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Joe Perches5c9f3022013-09-23 11:33:32 -0700527void tcp_send_probe0(struct sock *);
528void tcp_send_partial(struct sock *);
529int tcp_write_wakeup(struct sock *);
530void tcp_send_fin(struct sock *sk);
531void tcp_send_active_reset(struct sock *sk, gfp_t priority);
532int tcp_send_synack(struct sock *);
533bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
534 const char *proto);
535void tcp_push_one(struct sock *, unsigned int mss_now);
536void tcp_send_ack(struct sock *sk);
537void tcp_send_delayed_ack(struct sock *sk);
538void tcp_send_loss_probe(struct sock *sk);
539bool tcp_schedule_loss_probe(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
David S. Millera762a982005-07-05 15:18:51 -0700541/* tcp_input.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700542void tcp_resume_early_retransmit(struct sock *sk);
543void tcp_rearm_rto(struct sock *sk);
544void tcp_reset(struct sock *sk);
David S. Millera762a982005-07-05 15:18:51 -0700545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546/* tcp_timer.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700547void tcp_init_xmit_timers(struct sock *);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700548static inline void tcp_clear_xmit_timers(struct sock *sk)
549{
550 inet_csk_clear_xmit_timers(sk);
551}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Joe Perches5c9f3022013-09-23 11:33:32 -0700553unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
554unsigned int tcp_current_mss(struct sock *sk);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000555
556/* Bound MSS / TSO packet size with the half of the window */
557static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
558{
Alexey Kuznetsov01f83d62010-09-15 10:27:52 -0700559 int cutoff;
560
561 /* When peer uses tiny windows, there is no use in packetizing
562 * to sub-MSS pieces for the sake of SWS or making sure there
563 * are enough packets in the pipe for fast recovery.
564 *
565 * On the other hand, for extremely large MSS devices, handling
566 * smaller than MSS windows in this way does make sense.
567 */
568 if (tp->max_window >= 512)
569 cutoff = (tp->max_window >> 1);
570 else
571 cutoff = tp->max_window;
572
573 if (cutoff && pktsize > cutoff)
574 return max_t(int, cutoff, 68U - tp->tcp_header_len);
Ilpo Järvinen0c54b852009-03-14 14:23:05 +0000575 else
576 return pktsize;
577}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Arnaldo Carvalho de Melo17b085e2005-08-12 12:59:17 -0300579/* tcp.c */
Joe Perches5c9f3022013-09-23 11:33:32 -0700580void tcp_get_info(const struct sock *, struct tcp_info *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582/* Read 'sendfile()'-style from a TCP socket */
583typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
584 unsigned int, size_t);
Joe Perches5c9f3022013-09-23 11:33:32 -0700585int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
586 sk_read_actor_t recv_actor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
Joe Perches5c9f3022013-09-23 11:33:32 -0700588void tcp_initialize_rcv_mss(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Joe Perches5c9f3022013-09-23 11:33:32 -0700590int tcp_mtu_to_mss(struct sock *sk, int pmtu);
591int tcp_mss_to_mtu(struct sock *sk, int mss);
592void tcp_mtup_init(struct sock *sk);
593void tcp_init_buffer_space(struct sock *sk);
John Heffner5d424d52006-03-20 17:53:41 -0800594
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000595static inline void tcp_bound_rto(const struct sock *sk)
596{
597 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
598 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
599}
600
601static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
602{
Eric Dumazet740b0f12014-02-26 14:02:48 -0800603 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000604}
605
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800606static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
608 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
609 ntohl(TCP_FLAG_ACK) |
610 snd_wnd);
611}
612
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800613static inline void tcp_fast_path_on(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614{
615 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
616}
617
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700618static inline void tcp_fast_path_check(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700620 struct tcp_sock *tp = tcp_sk(sk);
621
David S. Millerb03efcf2005-07-08 14:57:23 -0700622 if (skb_queue_empty(&tp->out_of_order_queue) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 tp->rcv_wnd &&
624 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
625 !tp->urg_data)
626 tcp_fast_path_on(tp);
627}
628
Satoru SATOH0c266892009-05-04 11:11:01 -0700629/* Compute the actual rto_min value */
630static inline u32 tcp_rto_min(struct sock *sk)
631{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400632 const struct dst_entry *dst = __sk_dst_get(sk);
Satoru SATOH0c266892009-05-04 11:11:01 -0700633 u32 rto_min = TCP_RTO_MIN;
634
635 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
636 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
637 return rto_min;
638}
639
Eric Dumazet740b0f12014-02-26 14:02:48 -0800640static inline u32 tcp_rto_min_us(struct sock *sk)
641{
642 return jiffies_to_usecs(tcp_rto_min(sk));
643}
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645/* Compute the actual receive window we are currently advertising.
646 * Rcv_nxt can be after the window if our peer push more data
647 * than the offered window.
648 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800649static inline u32 tcp_receive_window(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
651 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
652
653 if (win < 0)
654 win = 0;
655 return (u32) win;
656}
657
658/* Choose a new window, without checks for shrinking, and without
659 * scaling applied to the result. The caller does these things
660 * if necessary. This is a "raw" window selection.
661 */
Joe Perches5c9f3022013-09-23 11:33:32 -0700662u32 __tcp_select_window(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Pavel Emelyanovee995282012-04-19 03:40:39 +0000664void tcp_send_window_probe(struct sock *sk);
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666/* TCP timestamps are only 32-bits, this causes a slight
667 * complication on 64-bit systems since we store a snapshot
Stephen Hemminger31f34262005-11-15 15:17:10 -0800668 * of jiffies in the buffer control blocks below. We decided
669 * to use only the low 32-bits of jiffies and hide the ugly
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 * casts with the following macro.
671 */
672#define tcp_time_stamp ((__u32)(jiffies))
673
Changli Gaoa3433f32010-06-12 14:01:43 +0000674#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
675
676#define TCPHDR_FIN 0x01
677#define TCPHDR_SYN 0x02
678#define TCPHDR_RST 0x04
679#define TCPHDR_PSH 0x08
680#define TCPHDR_ACK 0x10
681#define TCPHDR_URG 0x20
682#define TCPHDR_ECE 0x40
683#define TCPHDR_CWR 0x80
684
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -0800685/* This is what the send packet queuing engine uses to pass
Eric Dumazetf86586f2010-07-15 21:41:00 -0700686 * TCP per-packet control information to the transmission code.
687 * We also store the host-order sequence numbers in here too.
688 * This is 44 bytes if IPV6 is enabled.
689 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 */
691struct tcp_skb_cb {
692 union {
693 struct inet_skb_parm h4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000694#if IS_ENABLED(CONFIG_IPV6)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 struct inet6_skb_parm h6;
696#endif
697 } header; /* For incoming frames */
698 __u32 seq; /* Starting sequence number */
699 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
700 __u32 when; /* used to compute rtt's */
Eric Dumazet4de075e2011-09-27 13:25:05 -0400701 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 __u8 sacked; /* State flags for SACK/FACK. */
704#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
705#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
706#define TCPCB_LOST 0x04 /* SKB is lost */
707#define TCPCB_TAGBITS 0x07 /* All tag bits */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
709#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
710
Neal Cardwellf4f9f6e2012-04-16 07:08:06 +0000711 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
712 /* 1 byte hole */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 __u32 ack_seq; /* Sequence number ACK'd */
714};
715
716#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
717
Eric Dumazetbd14b1b2012-05-04 05:14:02 +0000718/* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
719 *
720 * If we receive a SYN packet with these bits set, it means a network is
721 * playing bad games with TOS bits. In order to avoid possible false congestion
722 * notifications, we disable TCP ECN negociation.
723 */
724static inline void
Hannes Frederic Sowa5d134f12013-01-05 16:10:48 +0000725TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
726 struct net *net)
Eric Dumazetbd14b1b2012-05-04 05:14:02 +0000727{
728 const struct tcphdr *th = tcp_hdr(skb);
729
Hannes Frederic Sowa5d134f12013-01-05 16:10:48 +0000730 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
Eric Dumazetbd14b1b2012-05-04 05:14:02 +0000731 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
732 inet_rsk(req)->ecn_ok = 1;
733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735/* Due to TSO, an SKB can be composed of multiple actual
736 * packets. To keep these tracked properly, we use this.
737 */
738static inline int tcp_skb_pcount(const struct sk_buff *skb)
739{
Herbert Xu79671682006-06-22 02:40:14 -0700740 return skb_shinfo(skb)->gso_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}
742
743/* This is valid iff tcp_skb_pcount() > 1. */
744static inline int tcp_skb_mss(const struct sk_buff *skb)
745{
Herbert Xu79671682006-06-22 02:40:14 -0700746 return skb_shinfo(skb)->gso_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700749/* Events passed to congestion control interface */
750enum tcp_ca_event {
751 CA_EVENT_TX_START, /* first transmit when no packets in flight */
752 CA_EVENT_CWND_RESTART, /* congestion window restart */
753 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700754 CA_EVENT_LOSS, /* loss timeout */
755 CA_EVENT_FAST_ACK, /* in sequence ack */
756 CA_EVENT_SLOW_ACK, /* other ack */
757};
758
759/*
760 * Interface for adding new TCP congestion control handlers
761 */
762#define TCP_CA_NAME_MAX 16
Stephen Hemminger3ff825b2006-11-09 16:32:06 -0800763#define TCP_CA_MAX 128
764#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
765
Stephen Hemminger164891a2007-04-23 22:26:16 -0700766#define TCP_CONG_NON_RESTRICTED 0x1
Stephen Hemminger164891a2007-04-23 22:26:16 -0700767
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700768struct tcp_congestion_ops {
769 struct list_head list;
Stephen Hemminger164891a2007-04-23 22:26:16 -0700770 unsigned long flags;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700771
772 /* initialize private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300773 void (*init)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700774 /* cleanup private data (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300775 void (*release)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700776
777 /* return slow start threshold (required) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300778 u32 (*ssthresh)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700779 /* do new cwnd calculation (required) */
Eric Dumazet24901552014-05-02 21:18:05 -0700780 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700781 /* call before changing ca_state (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300782 void (*set_state)(struct sock *sk, u8 new_state);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700783 /* call when cwnd event occurs (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300784 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700785 /* new value of cwnd after loss (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300786 u32 (*undo_cwnd)(struct sock *sk);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700787 /* hook for packet ack accounting (optional) */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700788 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300789 /* get info for inet_diag (optional) */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300790 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700791
792 char name[TCP_CA_NAME_MAX];
793 struct module *owner;
794};
795
Joe Perches5c9f3022013-09-23 11:33:32 -0700796int tcp_register_congestion_control(struct tcp_congestion_ops *type);
797void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700798
Joe Perches5c9f3022013-09-23 11:33:32 -0700799void tcp_init_congestion_control(struct sock *sk);
800void tcp_cleanup_congestion_control(struct sock *sk);
801int tcp_set_default_congestion_control(const char *name);
802void tcp_get_default_congestion_control(char *name);
803void tcp_get_available_congestion_control(char *buf, size_t len);
804void tcp_get_allowed_congestion_control(char *buf, size_t len);
805int tcp_set_allowed_congestion_control(char *allowed);
806int tcp_set_congestion_control(struct sock *sk, const char *name);
Yuchung Cheng9f9843a72013-10-31 11:07:31 -0700807int tcp_slow_start(struct tcp_sock *tp, u32 acked);
Joe Perches5c9f3022013-09-23 11:33:32 -0700808void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700809
Stephen Hemminger5f8ef482005-06-23 20:37:36 -0700810extern struct tcp_congestion_ops tcp_init_congestion_ops;
Joe Perches5c9f3022013-09-23 11:33:32 -0700811u32 tcp_reno_ssthresh(struct sock *sk);
Eric Dumazet24901552014-05-02 21:18:05 -0700812void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
David S. Millera8acfba2005-06-23 23:45:02 -0700813extern struct tcp_congestion_ops tcp_reno;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700814
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300815static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700816{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300817 struct inet_connection_sock *icsk = inet_csk(sk);
818
819 if (icsk->icsk_ca_ops->set_state)
820 icsk->icsk_ca_ops->set_state(sk, ca_state);
821 icsk->icsk_ca_state = ca_state;
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700822}
823
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300824static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700825{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300826 const struct inet_connection_sock *icsk = inet_csk(sk);
827
828 if (icsk->icsk_ca_ops->cwnd_event)
829 icsk->icsk_ca_ops->cwnd_event(sk, event);
Stephen Hemminger317a76f2005-06-23 12:19:55 -0700830}
831
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300832/* These functions determine how the current flow behaves in respect of SACK
833 * handling. SACK is negotiated with the peer, and therefore it can vary
834 * between different flows.
835 *
836 * tcp_is_sack - SACK enabled
837 * tcp_is_reno - No SACK
838 * tcp_is_fack - FACK enabled, implies SACK enabled
839 */
840static inline int tcp_is_sack(const struct tcp_sock *tp)
841{
842 return tp->rx_opt.sack_ok;
843}
844
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000845static inline bool tcp_is_reno(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300846{
847 return !tcp_is_sack(tp);
848}
849
Eric Dumazeta2a385d2012-05-16 23:15:34 +0000850static inline bool tcp_is_fack(const struct tcp_sock *tp)
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300851{
Vijay Subramanianab562222011-12-20 13:23:24 +0000852 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300853}
854
855static inline void tcp_enable_fack(struct tcp_sock *tp)
856{
Vijay Subramanianab562222011-12-20 13:23:24 +0000857 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300858}
859
Yuchung Chengeed530b2012-05-02 13:30:03 +0000860/* TCP early-retransmit (ER) is similar to but more conservative than
861 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
862 */
863static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
864{
865 tp->do_early_retrans = sysctl_tcp_early_retrans &&
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000866 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
867 sysctl_tcp_reordering == 3;
Yuchung Chengeed530b2012-05-02 13:30:03 +0000868}
869
870static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
871{
872 tp->do_early_retrans = 0;
873}
874
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300875static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
876{
877 return tp->sacked_out + tp->lost_out;
878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880/* This determines how many packets are "in the network" to the best
881 * of our knowledge. In many cases it is conservative, but where
882 * detailed information is available from the receiver (via SACK
883 * blocks etc.) we can make more aggressive calculations.
884 *
885 * Use this for decisions involving congestion control, use just
886 * tp->packets_out to determine if the send queue is empty or not.
887 *
888 * Read this equation as:
889 *
890 * "Packets sent once on transmission queue" MINUS
891 * "Packets left network, but not honestly ACKed yet" PLUS
892 * "Packets fast retransmitted"
893 */
Stephen Hemminger40efc6f2006-01-03 16:03:49 -0800894static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
Ilpo Järvinen83ae4082007-08-09 14:37:30 +0300896 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Ilpo Järvinen0b6a05c2009-09-15 01:30:10 -0700899#define TCP_INFINITE_SSTHRESH 0x7fffffff
900
901static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
902{
903 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
904}
905
Yuchung Cheng684bad12012-09-02 17:38:04 +0000906static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
907{
908 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
909 (1 << inet_csk(sk)->icsk_ca_state);
910}
911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
Yuchung Cheng684bad12012-09-02 17:38:04 +0000913 * The exception is cwnd reduction phase, when cwnd is decreasing towards
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 * ssthresh.
915 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300916static inline __u32 tcp_current_ssthresh(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300918 const struct tcp_sock *tp = tcp_sk(sk);
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400919
Yuchung Cheng684bad12012-09-02 17:38:04 +0000920 if (tcp_in_cwnd_reduction(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 return tp->snd_ssthresh;
922 else
923 return max(tp->snd_ssthresh,
924 ((tp->snd_cwnd >> 1) +
925 (tp->snd_cwnd >> 2)));
926}
927
Ilpo Järvinenb9c45952007-07-27 16:36:17 +0300928/* Use define here intentionally to get WARN_ON location shown at the caller */
929#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Joe Perches5c9f3022013-09-23 11:33:32 -0700931void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
932__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Neal Cardwell6b5a5c02011-11-21 17:15:14 +0000934/* The maximum number of MSS of available cwnd for which TSO defers
935 * sending if not using sysctl_tcp_tso_win_divisor.
936 */
937static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
938{
939 return 3;
940}
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942/* Slow start with delack produces 3 packets of burst, so that
John Heffnerdd9e0dd2008-04-15 15:26:39 -0700943 * it is safe "de facto". This will be the default - same as
944 * the default reordering threshold - but if reordering increases,
945 * we must be able to allow cwnd to burst at least this much in order
946 * to not pull it back when holes are filled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 */
948static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
949{
John Heffnerdd9e0dd2008-04-15 15:26:39 -0700950 return tp->reordering;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
952
Ilpo Järvinen90840de2007-12-31 04:48:41 -0800953/* Returns end sequence number of the receiver's advertised window */
954static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
955{
956 return tp->snd_una + tp->snd_wnd;
957}
Eric Dumazete114a712014-04-30 11:58:13 -0700958
959/* We follow the spirit of RFC2861 to validate cwnd but implement a more
960 * flexible approach. The RFC suggests cwnd should not be raised unless
Neal Cardwellca8a2262014-05-22 10:41:08 -0400961 * it was fully used previously. And that's exactly what we do in
962 * congestion avoidance mode. But in slow start we allow cwnd to grow
963 * as long as the application has used half the cwnd.
Eric Dumazete114a712014-04-30 11:58:13 -0700964 * Example :
965 * cwnd is 10 (IW10), but application sends 9 frames.
966 * We allow cwnd to reach 18 when all frames are ACKed.
967 * This check is safe because it's as aggressive as slow start which already
968 * risks 100% overshoot. The advantage is that we discourage application to
969 * either send more filler packets or data to artificially blow up the cwnd
970 * usage, and allow application-limited process to probe bw more aggressively.
Eric Dumazete114a712014-04-30 11:58:13 -0700971 */
Eric Dumazet24901552014-05-02 21:18:05 -0700972static inline bool tcp_is_cwnd_limited(const struct sock *sk)
Eric Dumazete114a712014-04-30 11:58:13 -0700973{
974 const struct tcp_sock *tp = tcp_sk(sk);
975
Neal Cardwellca8a2262014-05-22 10:41:08 -0400976 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
977 if (tp->snd_cwnd <= tp->snd_ssthresh)
978 return tp->snd_cwnd < 2 * tp->max_packets_out;
979
980 return tp->is_cwnd_limited;
Eric Dumazete114a712014-04-30 11:58:13 -0700981}
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800982
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700983static inline void tcp_check_probe_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Eric Dumazetcf533ea2011-10-21 05:22:42 -0400985 const struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700986 const struct inet_connection_sock *icsk = inet_csk(sk);
Ilpo Järvinen9e412ba2007-04-20 22:18:02 -0700987
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700988 if (!tp->packets_out && !icsk->icsk_pending)
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700989 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
990 icsk->icsk_rto, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991}
992
Hantzis Fotisee7537b2009-03-02 22:42:02 -0800993static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 tp->snd_wl1 = seq;
996}
997
Hantzis Fotisee7537b2009-03-02 22:42:02 -0800998static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
1000 tp->snd_wl1 = seq;
1001}
1002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003/*
1004 * Calculate(/check) TCP checksum
1005 */
Frederik Deweerdtba7808e2007-02-04 20:15:27 -08001006static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1007 __be32 daddr, __wsum base)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008{
1009 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1010}
1011
Al Virob51655b2006-11-14 21:40:42 -08001012static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013{
Herbert Xufb286bb2005-11-10 13:01:24 -08001014 return __skb_checksum_complete(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015}
1016
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001017static inline bool tcp_checksum_complete(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018{
Herbert Xu60476372007-04-09 11:59:39 -07001019 return !skb_csum_unnecessary(skb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 __tcp_checksum_complete(skb);
1021}
1022
1023/* Prequeue for VJ style copy to user, combined with checksumming. */
1024
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001025static inline void tcp_prequeue_init(struct tcp_sock *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
1027 tp->ucopy.task = NULL;
1028 tp->ucopy.len = 0;
1029 tp->ucopy.memory = 0;
1030 skb_queue_head_init(&tp->ucopy.prequeue);
Chris Leech97fc2f02006-05-23 17:55:33 -07001031#ifdef CONFIG_NET_DMA
1032 tp->ucopy.dma_chan = NULL;
1033 tp->ucopy.wakeup = 0;
1034 tp->ucopy.pinned_list = NULL;
1035 tp->ucopy.dma_cookie = 0;
1036#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037}
1038
Joe Perches5c9f3022013-09-23 11:33:32 -07001039bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041#undef STATE_TRACE
1042
1043#ifdef STATE_TRACE
1044static const char *statename[]={
1045 "Unused","Established","Syn Sent","Syn Recv",
1046 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1047 "Close Wait","Last ACK","Listen","Closing"
1048};
1049#endif
Joe Perches5c9f3022013-09-23 11:33:32 -07001050void tcp_set_state(struct sock *sk, int state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Joe Perches5c9f3022013-09-23 11:33:32 -07001052void tcp_done(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001054static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055{
1056 rx_opt->dsack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 rx_opt->num_sacks = 0;
1058}
1059
Joe Perches5c9f3022013-09-23 11:33:32 -07001060u32 tcp_default_init_rwnd(u32 mss);
Yuchung Cheng85f16522013-06-11 15:35:32 -07001061
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062/* Determine a window scaling and initial window to offer. */
Joe Perches5c9f3022013-09-23 11:33:32 -07001063void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1064 __u32 *window_clamp, int wscale_ok,
1065 __u8 *rcv_wscale, __u32 init_rcv_wnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067static inline int tcp_win_from_space(int space)
1068{
1069 return sysctl_tcp_adv_win_scale<=0 ?
1070 (space>>(-sysctl_tcp_adv_win_scale)) :
1071 space - (space>>sysctl_tcp_adv_win_scale);
1072}
1073
1074/* Note: caller must be prepared to deal with negative returns */
1075static inline int tcp_space(const struct sock *sk)
1076{
1077 return tcp_win_from_space(sk->sk_rcvbuf -
1078 atomic_read(&sk->sk_rmem_alloc));
1079}
1080
1081static inline int tcp_full_space(const struct sock *sk)
1082{
1083 return tcp_win_from_space(sk->sk_rcvbuf);
1084}
1085
Stephen Hemminger40efc6f2006-01-03 16:03:49 -08001086static inline void tcp_openreq_init(struct request_sock *req,
1087 struct tcp_options_received *rx_opt,
Octavian Purdilae0f802f2014-06-17 11:25:37 +03001088 struct sk_buff *skb, struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001090 struct inet_request_sock *ireq = inet_rsk(req);
1091
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
Florian Westphal4dfc2812008-04-10 03:12:40 -07001093 req->cookie_ts = 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001094 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
Jerry Chu10467162012-08-31 12:29:11 +00001095 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
Neal Cardwell016818d2012-09-22 04:18:55 +00001096 tcp_rsk(req)->snt_synack = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 req->mss = rx_opt->mss_clamp;
1098 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -07001099 ireq->tstamp_ok = rx_opt->tstamp_ok;
1100 ireq->sack_ok = rx_opt->sack_ok;
1101 ireq->snd_wscale = rx_opt->snd_wscale;
1102 ireq->wscale_ok = rx_opt->wscale_ok;
1103 ireq->acked = 0;
1104 ireq->ecn_ok = 0;
Eric Dumazet634fb9792013-10-09 15:21:29 -07001105 ireq->ir_rmt_port = tcp_hdr(skb)->source;
Eric Dumazetb44084c2013-10-10 00:04:37 -07001106 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
Octavian Purdilae0f802f2014-06-17 11:25:37 +03001107 ireq->ir_mark = inet_request_mark(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108}
1109
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001110extern void tcp_openreq_init_rwin(struct request_sock *req,
1111 struct sock *sk, struct dst_entry *dst);
1112
Joe Perches5c9f3022013-09-23 11:33:32 -07001113void tcp_enter_memory_pressure(struct sock *sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1116{
1117 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1118}
1119
1120static inline int keepalive_time_when(const struct tcp_sock *tp)
1121{
1122 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1123}
1124
Eric Dumazetdf19a622009-08-28 23:48:54 -07001125static inline int keepalive_probes(const struct tcp_sock *tp)
1126{
1127 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1128}
1129
Flavio Leitner6c37e5d2010-04-26 18:33:27 +00001130static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1131{
1132 const struct inet_connection_sock *icsk = &tp->inet_conn;
1133
1134 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1135 tcp_time_stamp - tp->rcv_tstamp);
1136}
1137
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001138static inline int tcp_fin_time(const struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001140 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1141 const int rto = inet_csk(sk)->icsk_rto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -07001143 if (fin_timeout < (rto << 2) - (rto >> 1))
1144 fin_timeout = (rto << 2) - (rto >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 return fin_timeout;
1147}
1148
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001149static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1150 int paws_win)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001152 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001153 return true;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001154 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001155 return true;
Eric Dumazetbc2ce892010-12-16 14:08:34 -08001156 /*
1157 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1158 * then following tcp messages have valid values. Ignore 0 value,
1159 * or else 'negative' tsval might forbid us to accept their packets.
1160 */
1161 if (!rx_opt->ts_recent)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001162 return true;
1163 return false;
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001164}
1165
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001166static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1167 int rst)
Ilpo Järvinenc887e6d2009-03-14 14:23:03 +00001168{
1169 if (tcp_paws_check(rx_opt, 0))
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001170 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 /* RST segments are not recommended to carry timestamp,
1173 and, if they do, it is recommended to ignore PAWS because
1174 "their cleanup function should take precedence over timestamps."
1175 Certainly, it is mistake. It is necessary to understand the reasons
1176 of this constraint to relax it: if peer reboots, clock may go
1177 out-of-sync and half-open connections will not be reset.
1178 Actually, the problem would be not existing if all
1179 the implementations followed draft about maintaining clock
1180 via reboots. Linux-2.2 DOES NOT!
1181
1182 However, we can relax time bounds for RST segments to MSL.
1183 */
James Morris9d729f72007-03-04 16:12:44 -08001184 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001185 return false;
1186 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
Pavel Emelyanova9c193292008-07-16 20:21:42 -07001189static inline void tcp_mib_init(struct net *net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190{
1191 /* See RFC 2012 */
Pavel Emelyanovcf1100a2008-07-16 20:27:38 -07001192 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1193 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1194 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1195 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196}
1197
Ilpo Järvinen5af4ec22007-09-20 11:30:48 -07001198/* from STCP */
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001199static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
David S. Miller0800f172007-09-20 11:40:37 -07001200{
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001201 tp->lost_skb_hint = NULL;
Ilpo Järvinenef9da472008-09-20 21:25:15 -07001202}
1203
1204static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1205{
1206 tcp_clear_retrans_hints_partial(tp);
Stephen Hemminger6a438bb2005-11-10 17:14:59 -08001207 tp->retransmit_skb_hint = NULL;
Ilpo Järvinenb7689202007-09-20 11:37:19 -07001208}
1209
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001210/* MD5 Signature */
1211struct crypto_hash;
1212
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001213union tcp_md5_addr {
1214 struct in_addr a4;
1215#if IS_ENABLED(CONFIG_IPV6)
1216 struct in6_addr a6;
1217#endif
1218};
1219
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001220/* - key database */
1221struct tcp_md5sig_key {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001222 struct hlist_node node;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001223 u8 keylen;
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001224 u8 family; /* AF_INET or AF_INET6 */
1225 union tcp_md5_addr addr;
1226 u8 key[TCP_MD5SIG_MAXKEYLEN];
1227 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001228};
1229
1230/* - sock block */
1231struct tcp_md5sig_info {
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001232 struct hlist_head head;
Eric Dumazeta8afca02012-01-31 18:45:40 +00001233 struct rcu_head rcu;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001234};
1235
1236/* - pseudo header */
1237struct tcp4_pseudohdr {
1238 __be32 saddr;
1239 __be32 daddr;
1240 __u8 pad;
1241 __u8 protocol;
1242 __be16 len;
1243};
1244
1245struct tcp6_pseudohdr {
1246 struct in6_addr saddr;
1247 struct in6_addr daddr;
1248 __be32 len;
1249 __be32 protocol; /* including padding */
1250};
1251
1252union tcp_md5sum_block {
1253 struct tcp4_pseudohdr ip4;
Eric Dumazetdfd56b82011-12-10 09:48:31 +00001254#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001255 struct tcp6_pseudohdr ip6;
1256#endif
1257};
1258
1259/* - pool: digest algorithm, hash description and scratch buffer */
1260struct tcp_md5sig_pool {
1261 struct hash_desc md5_desc;
1262 union tcp_md5sum_block md5_blk;
1263};
1264
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001265/* - functions */
Joe Perches5c9f3022013-09-23 11:33:32 -07001266int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1267 const struct sock *sk, const struct request_sock *req,
1268 const struct sk_buff *skb);
1269int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1270 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1271int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1272 int family);
1273struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001274 struct sock *addr_sk);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001275
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001276#ifdef CONFIG_TCP_MD5SIG
Joe Perches5c9f3022013-09-23 11:33:32 -07001277struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1278 const union tcp_md5_addr *addr,
1279 int family);
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001280#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001281#else
Eric Dumazeta915da9b2012-01-31 05:18:33 +00001282static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1283 const union tcp_md5_addr *addr,
1284 int family)
1285{
1286 return NULL;
1287}
YOSHIFUJI Hideaki9501f972008-04-18 12:45:16 +09001288#define tcp_twsk_md5_key(twsk) NULL
1289#endif
1290
Joe Perches5c9f3022013-09-23 11:33:32 -07001291bool tcp_alloc_md5sig_pool(void);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001292
Joe Perches5c9f3022013-09-23 11:33:32 -07001293struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
Eric Dumazet71cea172013-05-20 06:52:26 +00001294static inline void tcp_put_md5sig_pool(void)
1295{
1296 local_bh_enable();
1297}
Eric Dumazet35790c02010-05-16 00:34:04 -07001298
Joe Perches5c9f3022013-09-23 11:33:32 -07001299int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1300int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1301 unsigned int header_len);
1302int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1303 const struct tcp_md5sig_key *key);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001304
Jerry Chu10467162012-08-31 12:29:11 +00001305/* From tcp_fastopen.c */
Joe Perches5c9f3022013-09-23 11:33:32 -07001306void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1307 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1308 unsigned long *last_syn_loss);
1309void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1310 struct tcp_fastopen_cookie *cookie, bool syn_lost);
Yuchung Cheng783237e2012-07-19 06:43:07 +00001311struct tcp_fastopen_request {
1312 /* Fast Open cookie. Size 0 means a cookie request */
1313 struct tcp_fastopen_cookie cookie;
1314 struct msghdr *data; /* data in MSG_FASTOPEN */
Eric Dumazetf5ddcbb2014-02-20 10:09:18 -08001315 size_t size;
1316 int copied; /* queued in tcp_connect() */
Yuchung Cheng783237e2012-07-19 06:43:07 +00001317};
Yuchung Cheng783237e2012-07-19 06:43:07 +00001318void tcp_free_fastopen_req(struct tcp_sock *tp);
1319
Jerry Chu10467162012-08-31 12:29:11 +00001320extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1321int tcp_fastopen_reset_cipher(void *key, unsigned int len);
Yuchung Cheng843f4a52014-05-11 20:22:11 -07001322bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1323 struct request_sock *req,
1324 struct tcp_fastopen_cookie *foc,
1325 struct dst_entry *dst);
Hannes Frederic Sowa222e83d2013-10-19 21:48:58 +02001326void tcp_fastopen_init_key_once(bool publish);
Jerry Chu10467162012-08-31 12:29:11 +00001327#define TCP_FASTOPEN_KEY_LENGTH 16
1328
1329/* Fastopen key context */
1330struct tcp_fastopen_context {
Eric Dumazet7ae86392013-06-25 01:21:06 -07001331 struct crypto_cipher *tfm;
1332 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1333 struct rcu_head rcu;
Jerry Chu10467162012-08-31 12:29:11 +00001334};
1335
David S. Millerfe067e82007-03-07 12:12:44 -08001336/* write queue abstraction */
1337static inline void tcp_write_queue_purge(struct sock *sk)
1338{
1339 struct sk_buff *skb;
1340
1341 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
Hideo Aoki3ab224b2007-12-31 00:11:19 -08001342 sk_wmem_free_skb(sk, skb);
1343 sk_mem_reclaim(sk);
Ilpo Järvinen8818a9d2009-12-02 22:24:02 -08001344 tcp_clear_all_retrans_hints(tcp_sk(sk));
David S. Millerfe067e82007-03-07 12:12:44 -08001345}
1346
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001347static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001348{
David S. Millercd07a8e2008-09-23 00:50:13 -07001349 return skb_peek(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001350}
1351
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001352static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001353{
David S. Millercd07a8e2008-09-23 00:50:13 -07001354 return skb_peek_tail(&sk->sk_write_queue);
David S. Millerfe067e82007-03-07 12:12:44 -08001355}
1356
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001357static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1358 const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001359{
David S. Millercd07a8e2008-09-23 00:50:13 -07001360 return skb_queue_next(&sk->sk_write_queue, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001361}
1362
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001363static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1364 const struct sk_buff *skb)
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001365{
1366 return skb_queue_prev(&sk->sk_write_queue, skb);
1367}
1368
David S. Millerfe067e82007-03-07 12:12:44 -08001369#define tcp_for_write_queue(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001370 skb_queue_walk(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001371
1372#define tcp_for_write_queue_from(skb, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001373 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001374
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001375#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
David S. Millercd07a8e2008-09-23 00:50:13 -07001376 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
Ilpo Järvinen234b6862007-12-02 00:48:02 +02001377
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001378static inline struct sk_buff *tcp_send_head(const struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001379{
1380 return sk->sk_send_head;
1381}
1382
David S. Millercd07a8e2008-09-23 00:50:13 -07001383static inline bool tcp_skb_is_last(const struct sock *sk,
1384 const struct sk_buff *skb)
1385{
1386 return skb_queue_is_last(&sk->sk_write_queue, skb);
1387}
1388
Eric Dumazetcf533ea2011-10-21 05:22:42 -04001389static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
David S. Millerfe067e82007-03-07 12:12:44 -08001390{
David S. Millercd07a8e2008-09-23 00:50:13 -07001391 if (tcp_skb_is_last(sk, skb))
David S. Millerfe067e82007-03-07 12:12:44 -08001392 sk->sk_send_head = NULL;
David S. Millercd07a8e2008-09-23 00:50:13 -07001393 else
1394 sk->sk_send_head = tcp_write_queue_next(sk, skb);
David S. Millerfe067e82007-03-07 12:12:44 -08001395}
1396
1397static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1398{
1399 if (sk->sk_send_head == skb_unlinked)
1400 sk->sk_send_head = NULL;
1401}
1402
1403static inline void tcp_init_send_head(struct sock *sk)
1404{
1405 sk->sk_send_head = NULL;
1406}
1407
1408static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1409{
1410 __skb_queue_tail(&sk->sk_write_queue, skb);
1411}
1412
1413static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1414{
1415 __tcp_add_write_queue_tail(sk, skb);
1416
1417 /* Queue it, remembering where we must start sending. */
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001418 if (sk->sk_send_head == NULL) {
David S. Millerfe067e82007-03-07 12:12:44 -08001419 sk->sk_send_head = skb;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001420
1421 if (tcp_sk(sk)->highest_sack == NULL)
1422 tcp_sk(sk)->highest_sack = skb;
1423 }
David S. Millerfe067e82007-03-07 12:12:44 -08001424}
1425
1426static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1427{
1428 __skb_queue_head(&sk->sk_write_queue, skb);
1429}
1430
1431/* Insert buff after skb on the write queue of sk. */
1432static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1433 struct sk_buff *buff,
1434 struct sock *sk)
1435{
Gerrit Renker7de6c032008-04-14 00:05:09 -07001436 __skb_queue_after(&sk->sk_write_queue, skb, buff);
David S. Millerfe067e82007-03-07 12:12:44 -08001437}
1438
David S. Miller43f59c82008-09-21 21:28:51 -07001439/* Insert new before skb on the write queue of sk. */
David S. Millerfe067e82007-03-07 12:12:44 -08001440static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1441 struct sk_buff *skb,
1442 struct sock *sk)
1443{
David S. Miller43f59c82008-09-21 21:28:51 -07001444 __skb_queue_before(&sk->sk_write_queue, skb, new);
Ilpo Järvinen6e421412007-11-19 23:24:09 -08001445
1446 if (sk->sk_send_head == skb)
1447 sk->sk_send_head = new;
David S. Millerfe067e82007-03-07 12:12:44 -08001448}
1449
1450static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1451{
1452 __skb_unlink(skb, &sk->sk_write_queue);
1453}
1454
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001455static inline bool tcp_write_queue_empty(struct sock *sk)
David S. Millerfe067e82007-03-07 12:12:44 -08001456{
1457 return skb_queue_empty(&sk->sk_write_queue);
1458}
1459
Krishna Kumar12d50c42009-12-08 22:26:13 +00001460static inline void tcp_push_pending_frames(struct sock *sk)
1461{
1462 if (tcp_send_head(sk)) {
1463 struct tcp_sock *tp = tcp_sk(sk);
1464
1465 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1466 }
1467}
1468
Neal Cardwellecb97192012-02-27 17:52:52 -05001469/* Start sequence of the skb just after the highest skb with SACKed
1470 * bit, valid only if sacked_out > 0 or when the caller has ensured
1471 * validity by itself.
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001472 */
1473static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1474{
1475 if (!tp->sacked_out)
1476 return tp->snd_una;
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001477
1478 if (tp->highest_sack == NULL)
1479 return tp->snd_nxt;
1480
Ilpo Järvinena47e5a92007-11-15 19:41:46 -08001481 return TCP_SKB_CB(tp->highest_sack)->seq;
1482}
1483
Ilpo Järvinen6859d492007-12-02 00:48:06 +02001484static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1485{
1486 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1487 tcp_write_queue_next(sk, skb);
1488}
1489
1490static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1491{
1492 return tcp_sk(sk)->highest_sack;
1493}
1494
1495static inline void tcp_highest_sack_reset(struct sock *sk)
1496{
1497 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1498}
1499
1500/* Called when old skb is about to be deleted (to be combined with new skb) */
1501static inline void tcp_highest_sack_combine(struct sock *sk,
1502 struct sk_buff *old,
1503 struct sk_buff *new)
1504{
1505 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1506 tcp_sk(sk)->highest_sack = new;
1507}
1508
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001509/* Determines whether this is a thin stream (which may suffer from
1510 * increased latency). Used to trigger latency-reducing mechanisms.
1511 */
Eric Dumazeta2a385d2012-05-16 23:15:34 +00001512static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
Andreas Petlund5aa4b322010-02-18 02:45:45 +00001513{
1514 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1515}
1516
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517/* /proc */
1518enum tcp_seq_states {
1519 TCP_SEQ_STATE_LISTENING,
1520 TCP_SEQ_STATE_OPENREQ,
1521 TCP_SEQ_STATE_ESTABLISHED,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522};
1523
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001524int tcp_seq_open(struct inode *inode, struct file *file);
1525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526struct tcp_seq_afinfo {
Arjan van de Ven73cb88e2011-10-30 06:46:30 +00001527 char *name;
1528 sa_family_t family;
1529 const struct file_operations *seq_fops;
1530 struct seq_operations seq_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531};
1532
1533struct tcp_iter_state {
Denis V. Luneva4146b12008-04-13 22:11:14 -07001534 struct seq_net_private p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 sa_family_t family;
1536 enum tcp_seq_states state;
1537 struct sock *syn_wait_sk;
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06001538 int bucket, offset, sbucket, num;
1539 kuid_t uid;
Tom Herberta8b690f2010-06-07 00:43:42 -07001540 loff_t last_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541};
1542
Joe Perches5c9f3022013-09-23 11:33:32 -07001543int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1544void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001546extern struct request_sock_ops tcp_request_sock_ops;
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001547extern struct request_sock_ops tcp6_request_sock_ops;
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001548
Joe Perches5c9f3022013-09-23 11:33:32 -07001549void tcp_v4_destroy_sock(struct sock *sk);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001550
Eric Dumazet28be6e02013-10-18 10:36:17 -07001551struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
Joe Perches5c9f3022013-09-23 11:33:32 -07001552 netdev_features_t features);
1553struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1554int tcp_gro_complete(struct sk_buff *skb);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001555
Joe Perches5c9f3022013-09-23 11:33:32 -07001556void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
Herbert Xuf4c50d92006-06-22 03:02:40 -07001557
Eric Dumazetc9bee3b72013-07-22 20:27:07 -07001558static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1559{
1560 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1561}
1562
1563static inline bool tcp_stream_memory_free(const struct sock *sk)
1564{
1565 const struct tcp_sock *tp = tcp_sk(sk);
1566 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1567
1568 return notsent_bytes < tcp_notsent_lowat(tp);
1569}
1570
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001571#ifdef CONFIG_PROC_FS
Joe Perches5c9f3022013-09-23 11:33:32 -07001572int tcp4_proc_init(void);
1573void tcp4_proc_exit(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001574#endif
1575
Octavian Purdila5db92c92014-06-25 17:09:59 +03001576int tcp_rtx_synack(struct sock *sk, struct request_sock *req);
1577
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001578/* TCP af-specific functions */
1579struct tcp_sock_af_ops {
1580#ifdef CONFIG_TCP_MD5SIG
1581 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1582 struct sock *addr_sk);
1583 int (*calc_md5_hash) (char *location,
1584 struct tcp_md5sig_key *md5,
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001585 const struct sock *sk,
1586 const struct request_sock *req,
1587 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001588 int (*md5_parse) (struct sock *sk,
1589 char __user *optval,
1590 int optlen);
1591#endif
1592};
1593
1594struct tcp_request_sock_ops {
Octavian Purdila2aec4a22014-06-25 17:10:00 +03001595 u16 mss_clamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001596#ifdef CONFIG_TCP_MD5SIG
1597 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1598 struct request_sock *req);
John Dykstrae3afe7b2009-07-16 05:04:51 +00001599 int (*calc_md5_hash) (char *location,
1600 struct tcp_md5sig_key *md5,
Eric Dumazet318cf7a2011-10-24 02:46:04 -04001601 const struct sock *sk,
1602 const struct request_sock *req,
1603 const struct sk_buff *skb);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001604#endif
Octavian Purdila16bea702014-06-25 17:09:53 +03001605 void (*init_req)(struct request_sock *req, struct sock *sk,
1606 struct sk_buff *skb);
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001607#ifdef CONFIG_SYN_COOKIES
1608 __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb,
1609 __u16 *mss);
1610#endif
Octavian Purdilad94e0412014-06-25 17:09:55 +03001611 struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl,
1612 const struct request_sock *req,
1613 bool *strict);
Octavian Purdila936b8bd2014-06-25 17:09:57 +03001614 __u32 (*init_seq)(const struct sk_buff *skb);
Octavian Purdilad6274bd2014-06-25 17:09:58 +03001615 int (*send_synack)(struct sock *sk, struct dst_entry *dst,
1616 struct flowi *fl, struct request_sock *req,
1617 u16 queue_mapping, struct tcp_fastopen_cookie *foc);
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -08001618};
1619
Octavian Purdilafb7b37a2014-06-25 17:09:54 +03001620#ifdef CONFIG_SYN_COOKIES
1621static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1622 struct sock *sk, struct sk_buff *skb,
1623 __u16 *mss)
1624{
1625 return ops->cookie_init_seq(sk, skb, mss);
1626}
1627#else
1628static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1629 struct sock *sk, struct sk_buff *skb,
1630 __u16 *mss)
1631{
1632 return 0;
1633}
1634#endif
1635
Joe Perches5c9f3022013-09-23 11:33:32 -07001636int tcpv4_offload_init(void);
Daniel Borkmann28850dc2013-06-07 05:11:46 +00001637
Joe Perches5c9f3022013-09-23 11:33:32 -07001638void tcp_v4_init(void);
1639void tcp_init(void);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03001640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641#endif /* _TCP_H */