11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * INET An implementation of the TCP/IP protocol suite for the LINUX 31da177e4SLinus Torvalds * operating system. INET is implemented using the BSD Socket 41da177e4SLinus Torvalds * interface as the means of communication with the user level. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Definitions for the TCP module. 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Version: @(#)tcp.h 1.0.5 05/23/93 91da177e4SLinus Torvalds * 1002c30a84SJesper Juhl * Authors: Ross Biro 111da177e4SLinus Torvalds * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 121da177e4SLinus Torvalds * 131da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 141da177e4SLinus Torvalds * modify it under the terms of the GNU General Public License 151da177e4SLinus Torvalds * as published by the Free Software Foundation; either version 161da177e4SLinus Torvalds * 2 of the License, or (at your option) any later version. 171da177e4SLinus Torvalds */ 181da177e4SLinus Torvalds #ifndef _TCP_H 191da177e4SLinus Torvalds #define _TCP_H 201da177e4SLinus Torvalds 211da177e4SLinus Torvalds #define FASTRETRANS_DEBUG 1 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds #include <linux/list.h> 241da177e4SLinus Torvalds #include <linux/tcp.h> 25187f1882SPaul Gortmaker #include <linux/bug.h> 261da177e4SLinus Torvalds #include <linux/slab.h> 271da177e4SLinus Torvalds #include <linux/cache.h> 281da177e4SLinus Torvalds #include <linux/percpu.h> 29fb286bb2SHerbert Xu #include <linux/skbuff.h> 30cfb6eeb4SYOSHIFUJI Hideaki #include <linux/crypto.h> 31c6aefafbSGlenn Griffin #include <linux/cryptohash.h> 32435cf559SWilliam Allen Simpson #include <linux/kref.h> 33740b0f18SEric Dumazet #include <linux/ktime.h> 343f421baaSArnaldo Carvalho de Melo 353f421baaSArnaldo Carvalho de Melo #include <net/inet_connection_sock.h> 36295ff7edSArnaldo Carvalho de Melo #include <net/inet_timewait_sock.h> 3777d8bf9cSArnaldo Carvalho de Melo #include <net/inet_hashtables.h> 381da177e4SLinus Torvalds #include <net/checksum.h> 392e6599cbSArnaldo Carvalho de Melo #include <net/request_sock.h> 401da177e4SLinus Torvalds #include <net/sock.h> 411da177e4SLinus Torvalds #include <net/snmp.h> 421da177e4SLinus Torvalds #include <net/ip.h> 43c752f073SArnaldo Carvalho de Melo #include <net/tcp_states.h> 44bdf1ee5dSIlpo Järvinen #include <net/inet_ecn.h> 450c266898SSatoru SATOH #include <net/dst.h> 46c752f073SArnaldo Carvalho de Melo 471da177e4SLinus Torvalds #include <linux/seq_file.h> 48180d8cd9SGlauber Costa #include <linux/memcontrol.h> 491da177e4SLinus Torvalds 500f7ff927SArnaldo Carvalho de Melo extern struct inet_hashinfo tcp_hashinfo; 511da177e4SLinus Torvalds 52dd24c001SEric Dumazet extern struct percpu_counter tcp_orphan_count; 535c9f3023SJoe Perches void tcp_time_wait(struct sock *sk, int state, int timeo); 541da177e4SLinus Torvalds 551da177e4SLinus Torvalds #define MAX_TCP_HEADER (128 + MAX_HEADER) 5633ad798cSAdam Langley #define MAX_TCP_OPTION_SPACE 40 571da177e4SLinus Torvalds 581da177e4SLinus Torvalds /* 591da177e4SLinus Torvalds * Never offer a window over 32767 without using window scaling. Some 601da177e4SLinus Torvalds * poor stacks do signed 16bit maths! 611da177e4SLinus Torvalds */ 621da177e4SLinus Torvalds #define MAX_TCP_WINDOW 32767U 631da177e4SLinus Torvalds 641da177e4SLinus Torvalds /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ 651da177e4SLinus Torvalds #define TCP_MIN_MSS 88U 661da177e4SLinus Torvalds 675d424d5aSJohn Heffner /* The least MTU to use for probing */ 68dcd8fb85SFan Du #define TCP_BASE_MSS 1024 695d424d5aSJohn Heffner 7005cbc0dbSFan Du /* probing interval, default to 10 minutes as per RFC4821 */ 7105cbc0dbSFan Du #define TCP_PROBE_INTERVAL 600 7205cbc0dbSFan Du 736b58e0a5SFan Du /* Specify interval when tcp mtu probing will stop */ 746b58e0a5SFan Du #define TCP_PROBE_THRESHOLD 8 756b58e0a5SFan Du 761da177e4SLinus Torvalds /* After receiving this amount of duplicate ACKs fast retransmit starts. */ 771da177e4SLinus Torvalds #define TCP_FASTRETRANS_THRESH 3 781da177e4SLinus Torvalds 791da177e4SLinus Torvalds /* Maximal number of ACKs sent quickly to accelerate slow-start. */ 801da177e4SLinus Torvalds #define TCP_MAX_QUICKACKS 16U 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds /* urg_data states */ 831da177e4SLinus Torvalds #define TCP_URG_VALID 0x0100 841da177e4SLinus Torvalds #define TCP_URG_NOTYET 0x0200 851da177e4SLinus Torvalds #define TCP_URG_READ 0x0400 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds #define TCP_RETR1 3 /* 881da177e4SLinus Torvalds * This is how many retries it does before it 891da177e4SLinus Torvalds * tries to figure out if the gateway is 901da177e4SLinus Torvalds * down. Minimal RFC value is 3; it corresponds 911da177e4SLinus Torvalds * to ~3sec-8min depending on RTO. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds #define TCP_RETR2 15 /* 951da177e4SLinus Torvalds * This should take at least 961da177e4SLinus Torvalds * 90 minutes to time out. 971da177e4SLinus Torvalds * RFC1122 says that the limit is 100 sec. 981da177e4SLinus Torvalds * 15 is ~13-30min depending on RTO. 991da177e4SLinus Torvalds */ 1001da177e4SLinus Torvalds 1016c9ff979SAlex Bergmann #define TCP_SYN_RETRIES 6 /* This is how many retries are done 1026c9ff979SAlex Bergmann * when active opening a connection. 1036c9ff979SAlex Bergmann * RFC1122 says the minimum retry MUST 1046c9ff979SAlex Bergmann * be at least 180secs. Nevertheless 1056c9ff979SAlex Bergmann * this value is corresponding to 1066c9ff979SAlex Bergmann * 63secs of retransmission with the 1076c9ff979SAlex Bergmann * current initial RTO. 1086c9ff979SAlex Bergmann */ 1091da177e4SLinus Torvalds 1106c9ff979SAlex Bergmann #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done 1116c9ff979SAlex Bergmann * when passive opening a connection. 1126c9ff979SAlex Bergmann * This is corresponding to 31secs of 1136c9ff979SAlex Bergmann * retransmission with the current 1146c9ff979SAlex Bergmann * initial RTO. 1156c9ff979SAlex Bergmann */ 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT 1181da177e4SLinus Torvalds * state, about 60 seconds */ 1191da177e4SLinus Torvalds #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN 1201da177e4SLinus Torvalds /* BSD style FIN_WAIT2 deadlock breaker. 1211da177e4SLinus Torvalds * It used to be 3min, new value is 60sec, 1221da177e4SLinus Torvalds * to combine FIN-WAIT-2 timeout with 1231da177e4SLinus Torvalds * TIME-WAIT timer. 1241da177e4SLinus Torvalds */ 1251da177e4SLinus Torvalds 1261da177e4SLinus Torvalds #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ 1271da177e4SLinus Torvalds #if HZ >= 100 1281da177e4SLinus Torvalds #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ 1291da177e4SLinus Torvalds #define TCP_ATO_MIN ((unsigned)(HZ/25)) 1301da177e4SLinus Torvalds #else 1311da177e4SLinus Torvalds #define TCP_DELACK_MIN 4U 1321da177e4SLinus Torvalds #define TCP_ATO_MIN 4U 1331da177e4SLinus Torvalds #endif 1341da177e4SLinus Torvalds #define TCP_RTO_MAX ((unsigned)(120*HZ)) 1351da177e4SLinus Torvalds #define TCP_RTO_MIN ((unsigned)(HZ/5)) 136fd4f2ceaSEric Dumazet #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ 1379ad7c049SJerry Chu #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now 1389ad7c049SJerry Chu * used as a fallback RTO for the 1399ad7c049SJerry Chu * initial data transmission if no 1409ad7c049SJerry Chu * valid RTT sample has been acquired, 1419ad7c049SJerry Chu * most likely due to retrans in 3WHS. 1429ad7c049SJerry Chu */ 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes 1451da177e4SLinus Torvalds * for local resources. 1461da177e4SLinus Torvalds */ 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ 1491da177e4SLinus Torvalds #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ 1501da177e4SLinus Torvalds #define TCP_KEEPALIVE_INTVL (75*HZ) 1511da177e4SLinus Torvalds 1521da177e4SLinus Torvalds #define MAX_TCP_KEEPIDLE 32767 1531da177e4SLinus Torvalds #define MAX_TCP_KEEPINTVL 32767 1541da177e4SLinus Torvalds #define MAX_TCP_KEEPCNT 127 1551da177e4SLinus Torvalds #define MAX_TCP_SYNCNT 127 1561da177e4SLinus Torvalds 1571da177e4SLinus Torvalds #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) 1601da177e4SLinus Torvalds #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated 1611da177e4SLinus Torvalds * after this time. It should be equal 1621da177e4SLinus Torvalds * (or greater than) TCP_TIMEWAIT_LEN 1631da177e4SLinus Torvalds * to provide reliability equal to one 1641da177e4SLinus Torvalds * provided by timewait state. 1651da177e4SLinus Torvalds */ 1661da177e4SLinus Torvalds #define TCP_PAWS_WINDOW 1 /* Replay window for per-host 1671da177e4SLinus Torvalds * timestamps. It must be less than 1681da177e4SLinus Torvalds * minimal timewait lifetime. 1691da177e4SLinus Torvalds */ 1701da177e4SLinus Torvalds /* 1711da177e4SLinus Torvalds * TCP option 1721da177e4SLinus Torvalds */ 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds #define TCPOPT_NOP 1 /* Padding */ 1751da177e4SLinus Torvalds #define TCPOPT_EOL 0 /* End of options */ 1761da177e4SLinus Torvalds #define TCPOPT_MSS 2 /* Segment size negotiating */ 1771da177e4SLinus Torvalds #define TCPOPT_WINDOW 3 /* Window scaling */ 1781da177e4SLinus Torvalds #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ 1791da177e4SLinus Torvalds #define TCPOPT_SACK 5 /* SACK Block */ 1801da177e4SLinus Torvalds #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ 181cfb6eeb4SYOSHIFUJI Hideaki #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ 1827f9b838bSDaniel Lee #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ 1832100c8d2SYuchung Cheng #define TCPOPT_EXP 254 /* Experimental */ 1842100c8d2SYuchung Cheng /* Magic number to be after the option value for sharing TCP 1852100c8d2SYuchung Cheng * experimental options. See draft-ietf-tcpm-experimental-options-00.txt 1862100c8d2SYuchung Cheng */ 1872100c8d2SYuchung Cheng #define TCPOPT_FASTOPEN_MAGIC 0xF989 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds /* 1901da177e4SLinus Torvalds * TCP option lengths 1911da177e4SLinus Torvalds */ 1921da177e4SLinus Torvalds 1931da177e4SLinus Torvalds #define TCPOLEN_MSS 4 1941da177e4SLinus Torvalds #define TCPOLEN_WINDOW 3 1951da177e4SLinus Torvalds #define TCPOLEN_SACK_PERM 2 1961da177e4SLinus Torvalds #define TCPOLEN_TIMESTAMP 10 197cfb6eeb4SYOSHIFUJI Hideaki #define TCPOLEN_MD5SIG 18 1987f9b838bSDaniel Lee #define TCPOLEN_FASTOPEN_BASE 2 1992100c8d2SYuchung Cheng #define TCPOLEN_EXP_FASTOPEN_BASE 4 2001da177e4SLinus Torvalds 2011da177e4SLinus Torvalds /* But this is what stacks really send out. */ 2021da177e4SLinus Torvalds #define TCPOLEN_TSTAMP_ALIGNED 12 2031da177e4SLinus Torvalds #define TCPOLEN_WSCALE_ALIGNED 4 2041da177e4SLinus Torvalds #define TCPOLEN_SACKPERM_ALIGNED 4 2051da177e4SLinus Torvalds #define TCPOLEN_SACK_BASE 2 2061da177e4SLinus Torvalds #define TCPOLEN_SACK_BASE_ALIGNED 4 2071da177e4SLinus Torvalds #define TCPOLEN_SACK_PERBLOCK 8 208cfb6eeb4SYOSHIFUJI Hideaki #define TCPOLEN_MD5SIG_ALIGNED 20 20933ad798cSAdam Langley #define TCPOLEN_MSS_ALIGNED 4 2101da177e4SLinus Torvalds 2111da177e4SLinus Torvalds /* Flags in tp->nonagle */ 2121da177e4SLinus Torvalds #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 2131da177e4SLinus Torvalds #define TCP_NAGLE_CORK 2 /* Socket is corked */ 214caa20d9aSStephen Hemminger #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ 2151da177e4SLinus Torvalds 21636e31b0aSAndreas Petlund /* TCP thin-stream limits */ 21736e31b0aSAndreas Petlund #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ 21836e31b0aSAndreas Petlund 2197eb38527SDavid S. Miller /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */ 220442b9635SDavid S. Miller #define TCP_INIT_CWND 10 221442b9635SDavid S. Miller 222cf60af03SYuchung Cheng /* Bit Flags for sysctl_tcp_fastopen */ 223cf60af03SYuchung Cheng #define TFO_CLIENT_ENABLE 1 22410467163SJerry Chu #define TFO_SERVER_ENABLE 2 22567da22d2SYuchung Cheng #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ 226cf60af03SYuchung Cheng 22710467163SJerry Chu /* Accept SYN data w/o any cookie option */ 22810467163SJerry Chu #define TFO_SERVER_COOKIE_NOT_REQD 0x200 22910467163SJerry Chu 23010467163SJerry Chu /* Force enable TFO on all listeners, i.e., not requiring the 23110467163SJerry Chu * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. 23210467163SJerry Chu */ 23310467163SJerry Chu #define TFO_SERVER_WO_SOCKOPT1 0x400 23410467163SJerry Chu #define TFO_SERVER_WO_SOCKOPT2 0x800 23510467163SJerry Chu 236295ff7edSArnaldo Carvalho de Melo extern struct inet_timewait_death_row tcp_death_row; 237295ff7edSArnaldo Carvalho de Melo 2381da177e4SLinus Torvalds /* sysctl variables for tcp */ 2391da177e4SLinus Torvalds extern int sysctl_tcp_timestamps; 2401da177e4SLinus Torvalds extern int sysctl_tcp_window_scaling; 2411da177e4SLinus Torvalds extern int sysctl_tcp_sack; 2421da177e4SLinus Torvalds extern int sysctl_tcp_fin_timeout; 2431da177e4SLinus Torvalds extern int sysctl_tcp_keepalive_time; 2441da177e4SLinus Torvalds extern int sysctl_tcp_keepalive_probes; 2451da177e4SLinus Torvalds extern int sysctl_tcp_keepalive_intvl; 2461da177e4SLinus Torvalds extern int sysctl_tcp_syn_retries; 2471da177e4SLinus Torvalds extern int sysctl_tcp_synack_retries; 2481da177e4SLinus Torvalds extern int sysctl_tcp_retries1; 2491da177e4SLinus Torvalds extern int sysctl_tcp_retries2; 2501da177e4SLinus Torvalds extern int sysctl_tcp_orphan_retries; 2511da177e4SLinus Torvalds extern int sysctl_tcp_syncookies; 2522100c8d2SYuchung Cheng extern int sysctl_tcp_fastopen; 2531da177e4SLinus Torvalds extern int sysctl_tcp_retrans_collapse; 2541da177e4SLinus Torvalds extern int sysctl_tcp_stdurg; 2551da177e4SLinus Torvalds extern int sysctl_tcp_rfc1337; 2561da177e4SLinus Torvalds extern int sysctl_tcp_abort_on_overflow; 2571da177e4SLinus Torvalds extern int sysctl_tcp_max_orphans; 2581da177e4SLinus Torvalds extern int sysctl_tcp_fack; 2591da177e4SLinus Torvalds extern int sysctl_tcp_reordering; 260dca145ffSEric Dumazet extern int sysctl_tcp_max_reordering; 2611da177e4SLinus Torvalds extern int sysctl_tcp_dsack; 262a4fe34bfSEric W. Biederman extern long sysctl_tcp_mem[3]; 2631da177e4SLinus Torvalds extern int sysctl_tcp_wmem[3]; 2641da177e4SLinus Torvalds extern int sysctl_tcp_rmem[3]; 2651da177e4SLinus Torvalds extern int sysctl_tcp_app_win; 2661da177e4SLinus Torvalds extern int sysctl_tcp_adv_win_scale; 2671da177e4SLinus Torvalds extern int sysctl_tcp_tw_reuse; 2681da177e4SLinus Torvalds extern int sysctl_tcp_frto; 2691da177e4SLinus Torvalds extern int sysctl_tcp_low_latency; 2701da177e4SLinus Torvalds extern int sysctl_tcp_nometrics_save; 2711da177e4SLinus Torvalds extern int sysctl_tcp_moderate_rcvbuf; 2721da177e4SLinus Torvalds extern int sysctl_tcp_tso_win_divisor; 27315d99e02SRick Jones extern int sysctl_tcp_workaround_signed_windows; 27435089bb2SDavid S. Miller extern int sysctl_tcp_slow_start_after_idle; 27536e31b0aSAndreas Petlund extern int sysctl_tcp_thin_linear_timeouts; 2767e380175SAndreas Petlund extern int sysctl_tcp_thin_dupack; 277eed530b6SYuchung Cheng extern int sysctl_tcp_early_retrans; 27846d3ceabSEric Dumazet extern int sysctl_tcp_limit_output_bytes; 279282f23c6SEric Dumazet extern int sysctl_tcp_challenge_ack_limit; 280c9bee3b7SEric Dumazet extern unsigned int sysctl_tcp_notsent_lowat; 28195bd09ebSEric Dumazet extern int sysctl_tcp_min_tso_segs; 282f54b3111SEric Dumazet extern int sysctl_tcp_autocorking; 283032ee423SNeal Cardwell extern int sysctl_tcp_invalid_ratelimit; 2841da177e4SLinus Torvalds 2858d987e5cSEric Dumazet extern atomic_long_t tcp_memory_allocated; 2861748376bSEric Dumazet extern struct percpu_counter tcp_sockets_allocated; 2871da177e4SLinus Torvalds extern int tcp_memory_pressure; 2881da177e4SLinus Torvalds 2891da177e4SLinus Torvalds /* 2901da177e4SLinus Torvalds * The next routines deal with comparing 32 bit unsigned ints 2911da177e4SLinus Torvalds * and worry about wraparound (automatic with unsigned arithmetic). 2921da177e4SLinus Torvalds */ 2931da177e4SLinus Torvalds 294a2a385d6SEric Dumazet static inline bool before(__u32 seq1, __u32 seq2) 2951da177e4SLinus Torvalds { 2960d630cc0SGerrit Renker return (__s32)(seq1-seq2) < 0; 2971da177e4SLinus Torvalds } 2989a036b9cSGerrit Renker #define after(seq2, seq1) before(seq1, seq2) 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds /* is s2<=s1<=s3 ? */ 301a2a385d6SEric Dumazet static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) 3021da177e4SLinus Torvalds { 3031da177e4SLinus Torvalds return seq3 - seq2 >= seq1 - seq2; 3041da177e4SLinus Torvalds } 3051da177e4SLinus Torvalds 306efcdbf24SArun Sharma static inline bool tcp_out_of_memory(struct sock *sk) 307efcdbf24SArun Sharma { 308efcdbf24SArun Sharma if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && 309efcdbf24SArun Sharma sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) 310efcdbf24SArun Sharma return true; 311efcdbf24SArun Sharma return false; 312efcdbf24SArun Sharma } 313efcdbf24SArun Sharma 314ad1af0feSDavid S. Miller static inline bool tcp_too_many_orphans(struct sock *sk, int shift) 315e4fd5da3SPavel Emelianov { 316ad1af0feSDavid S. Miller struct percpu_counter *ocp = sk->sk_prot->orphan_count; 317ad1af0feSDavid S. Miller int orphans = percpu_counter_read_positive(ocp); 318ad1af0feSDavid S. Miller 319ad1af0feSDavid S. Miller if (orphans << shift > sysctl_tcp_max_orphans) { 320ad1af0feSDavid S. Miller orphans = percpu_counter_sum_positive(ocp); 321ad1af0feSDavid S. Miller if (orphans << shift > sysctl_tcp_max_orphans) 322ad1af0feSDavid S. Miller return true; 323ad1af0feSDavid S. Miller } 324ad1af0feSDavid S. Miller return false; 325e4fd5da3SPavel Emelianov } 3261da177e4SLinus Torvalds 3275c9f3023SJoe Perches bool tcp_check_oom(struct sock *sk, int shift); 328efcdbf24SArun Sharma 329a0f82f64SFlorian Westphal /* syncookies: remember time of last synqueue overflow */ 330a0f82f64SFlorian Westphal static inline void tcp_synq_overflow(struct sock *sk) 331a0f82f64SFlorian Westphal { 332a0f82f64SFlorian Westphal tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies; 333a0f82f64SFlorian Westphal } 334a0f82f64SFlorian Westphal 335a0f82f64SFlorian Westphal /* syncookies: no recent synqueue overflow on this listening socket? */ 336a2a385d6SEric Dumazet static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) 337a0f82f64SFlorian Westphal { 338a0f82f64SFlorian Westphal unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; 3399ad7c049SJerry Chu return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK); 340a0f82f64SFlorian Westphal } 341a0f82f64SFlorian Westphal 3421da177e4SLinus Torvalds extern struct proto tcp_prot; 3431da177e4SLinus Torvalds 34457ef42d5SPavel Emelyanov #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) 34557ef42d5SPavel Emelyanov #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field) 34657ef42d5SPavel Emelyanov #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) 34757ef42d5SPavel Emelyanov #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val) 348aa2ea058STom Herbert #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) 3491da177e4SLinus Torvalds 3505c9f3023SJoe Perches void tcp_tasklet_init(void); 35146d3ceabSEric Dumazet 3525c9f3023SJoe Perches void tcp_v4_err(struct sk_buff *skb, u32); 3531da177e4SLinus Torvalds 3545c9f3023SJoe Perches void tcp_shutdown(struct sock *sk, int how); 3551da177e4SLinus Torvalds 3565c9f3023SJoe Perches void tcp_v4_early_demux(struct sk_buff *skb); 3575c9f3023SJoe Perches int tcp_v4_rcv(struct sk_buff *skb); 3581da177e4SLinus Torvalds 3595c9f3023SJoe Perches int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); 3601b784140SYing Xue int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); 3615c9f3023SJoe Perches int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, 3625c9f3023SJoe Perches int flags); 3635c9f3023SJoe Perches void tcp_release_cb(struct sock *sk); 3645c9f3023SJoe Perches void tcp_wfree(struct sk_buff *skb); 3655c9f3023SJoe Perches void tcp_write_timer_handler(struct sock *sk); 3665c9f3023SJoe Perches void tcp_delack_timer_handler(struct sock *sk); 3675c9f3023SJoe Perches int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); 3685c9f3023SJoe Perches int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb, 369cf533ea5SEric Dumazet const struct tcphdr *th, unsigned int len); 3705c9f3023SJoe Perches void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, 371cf533ea5SEric Dumazet const struct tcphdr *th, unsigned int len); 3725c9f3023SJoe Perches void tcp_rcv_space_adjust(struct sock *sk); 3735c9f3023SJoe Perches int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); 3745c9f3023SJoe Perches void tcp_twsk_destructor(struct sock *sk); 3755c9f3023SJoe Perches ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, 37653d3176bSChangli Gao struct pipe_inode_info *pipe, size_t len, 37753d3176bSChangli Gao unsigned int flags); 3789c55e01cSJens Axboe 379463c84b9SArnaldo Carvalho de Melo static inline void tcp_dec_quickack_mode(struct sock *sk, 380463c84b9SArnaldo Carvalho de Melo const unsigned int pkts) 3811da177e4SLinus Torvalds { 382463c84b9SArnaldo Carvalho de Melo struct inet_connection_sock *icsk = inet_csk(sk); 383fc6415bcSDavid S. Miller 384463c84b9SArnaldo Carvalho de Melo if (icsk->icsk_ack.quick) { 385463c84b9SArnaldo Carvalho de Melo if (pkts >= icsk->icsk_ack.quick) { 386463c84b9SArnaldo Carvalho de Melo icsk->icsk_ack.quick = 0; 3871da177e4SLinus Torvalds /* Leaving quickack mode we deflate ATO. */ 388463c84b9SArnaldo Carvalho de Melo icsk->icsk_ack.ato = TCP_ATO_MIN; 389fc6415bcSDavid S. Miller } else 390463c84b9SArnaldo Carvalho de Melo icsk->icsk_ack.quick -= pkts; 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 394bdf1ee5dSIlpo Järvinen #define TCP_ECN_OK 1 395bdf1ee5dSIlpo Järvinen #define TCP_ECN_QUEUE_CWR 2 396bdf1ee5dSIlpo Järvinen #define TCP_ECN_DEMAND_CWR 4 3977a269ffaSEric Dumazet #define TCP_ECN_SEEN 8 398bdf1ee5dSIlpo Järvinen 399fd2c3ef7SEric Dumazet enum tcp_tw_status { 4001da177e4SLinus Torvalds TCP_TW_SUCCESS = 0, 4011da177e4SLinus Torvalds TCP_TW_RST = 1, 4021da177e4SLinus Torvalds TCP_TW_ACK = 2, 4031da177e4SLinus Torvalds TCP_TW_SYN = 3 4041da177e4SLinus Torvalds }; 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds 4075c9f3023SJoe Perches enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, 4081da177e4SLinus Torvalds struct sk_buff *skb, 4098feaf0c0SArnaldo Carvalho de Melo const struct tcphdr *th); 4105c9f3023SJoe Perches struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, 41152452c54SEric Dumazet struct request_sock *req, bool fastopen); 4125c9f3023SJoe Perches int tcp_child_process(struct sock *parent, struct sock *child, 4131da177e4SLinus Torvalds struct sk_buff *skb); 4145ae344c9SNeal Cardwell void tcp_enter_loss(struct sock *sk); 4155c9f3023SJoe Perches void tcp_clear_retrans(struct tcp_sock *tp); 4165c9f3023SJoe Perches void tcp_update_metrics(struct sock *sk); 4175c9f3023SJoe Perches void tcp_init_metrics(struct sock *sk); 4185c9f3023SJoe Perches void tcp_metrics_init(void); 4195c9f3023SJoe Perches bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, 420a26552afSHannes Frederic Sowa bool paws_check, bool timestamps); 4215c9f3023SJoe Perches bool tcp_remember_stamp(struct sock *sk); 4225c9f3023SJoe Perches bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); 4235c9f3023SJoe Perches void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); 4245c9f3023SJoe Perches void tcp_disable_fack(struct tcp_sock *tp); 4255c9f3023SJoe Perches void tcp_close(struct sock *sk, long timeout); 4265c9f3023SJoe Perches void tcp_init_sock(struct sock *sk); 4275c9f3023SJoe Perches unsigned int tcp_poll(struct file *file, struct socket *sock, 42853d3176bSChangli Gao struct poll_table_struct *wait); 4295c9f3023SJoe Perches int tcp_getsockopt(struct sock *sk, int level, int optname, 4303fdadf7dSDmitry Mishin char __user *optval, int __user *optlen); 4315c9f3023SJoe Perches int tcp_setsockopt(struct sock *sk, int level, int optname, 43253d3176bSChangli Gao char __user *optval, unsigned int optlen); 4335c9f3023SJoe Perches int compat_tcp_getsockopt(struct sock *sk, int level, int optname, 43453d3176bSChangli Gao char __user *optval, int __user *optlen); 4355c9f3023SJoe Perches int compat_tcp_setsockopt(struct sock *sk, int level, int optname, 436b7058842SDavid S. Miller char __user *optval, unsigned int optlen); 4375c9f3023SJoe Perches void tcp_set_keepalive(struct sock *sk, int val); 43842cb80a2SEric Dumazet void tcp_syn_ack_timeout(const struct request_sock *req); 4391b784140SYing Xue int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, 4401b784140SYing Xue int flags, int *addr_len); 4415c9f3023SJoe Perches void tcp_parse_options(const struct sk_buff *skb, 4421a2c6181SChristoph Paasch struct tcp_options_received *opt_rx, 4432100c8d2SYuchung Cheng int estab, struct tcp_fastopen_cookie *foc); 4445c9f3023SJoe Perches const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); 4457d5d5525SYOSHIFUJI Hideaki 4461da177e4SLinus Torvalds /* 4471da177e4SLinus Torvalds * TCP v4 functions exported for the inet6 API 4481da177e4SLinus Torvalds */ 4491da177e4SLinus Torvalds 4505c9f3023SJoe Perches void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); 4514fab9071SNeal Cardwell void tcp_v4_mtu_reduced(struct sock *sk); 45226e37360SEric Dumazet void tcp_req_err(struct sock *sk, u32 seq); 4535c9f3023SJoe Perches int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); 4545c9f3023SJoe Perches struct sock *tcp_create_openreq_child(struct sock *sk, 45560236fddSArnaldo Carvalho de Melo struct request_sock *req, 4561da177e4SLinus Torvalds struct sk_buff *skb); 45781164413SDaniel Borkmann void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); 4585c9f3023SJoe Perches struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, 45960236fddSArnaldo Carvalho de Melo struct request_sock *req, 4601da177e4SLinus Torvalds struct dst_entry *dst); 4615c9f3023SJoe Perches int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); 4625c9f3023SJoe Perches int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); 4635c9f3023SJoe Perches int tcp_connect(struct sock *sk); 4645c9f3023SJoe Perches struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst, 465e6b4d113SWilliam Allen Simpson struct request_sock *req, 4668336886fSJerry Chu struct tcp_fastopen_cookie *foc); 4675c9f3023SJoe Perches int tcp_disconnect(struct sock *sk, int flags); 4681da177e4SLinus Torvalds 469370816aeSPavel Emelyanov void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); 470292e8d8cSPavel Emelyanov int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); 47163d02d15SEric Dumazet void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); 4721da177e4SLinus Torvalds 4731da177e4SLinus Torvalds /* From syncookies.c */ 4745c9f3023SJoe Perches int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 4750198230bSPatrick McHardy u32 cookie); 476461b74c3SCong Wang struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); 477e05c82d3SEric Dumazet #ifdef CONFIG_SYN_COOKIES 4788c27bd75SFlorian Westphal 47963262315SEric Dumazet /* Syncookies use a monotonic timer which increments every 60 seconds. 4808c27bd75SFlorian Westphal * This counter is used both as a hash input and partially encoded into 4818c27bd75SFlorian Westphal * the cookie value. A cookie is only validated further if the delta 4828c27bd75SFlorian Westphal * between the current counter value and the encoded one is less than this, 48363262315SEric Dumazet * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if 4848c27bd75SFlorian Westphal * the counter advances immediately after a cookie is generated). 4858c27bd75SFlorian Westphal */ 4868c27bd75SFlorian Westphal #define MAX_SYNCOOKIE_AGE 2 4878c27bd75SFlorian Westphal 4888c27bd75SFlorian Westphal static inline u32 tcp_cookie_time(void) 4898c27bd75SFlorian Westphal { 49063262315SEric Dumazet u64 val = get_jiffies_64(); 49163262315SEric Dumazet 49263262315SEric Dumazet do_div(val, 60 * HZ); 49363262315SEric Dumazet return val; 4948c27bd75SFlorian Westphal } 4958c27bd75SFlorian Westphal 4965c9f3023SJoe Perches u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 4975c9f3023SJoe Perches u16 *mssp); 49857b47553SOctavian Purdila __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 49957b47553SOctavian Purdila __u16 *mss); 5005c9f3023SJoe Perches __u32 cookie_init_timestamp(struct request_sock *req); 501f1673381SFlorian Westphal bool cookie_timestamp_decode(struct tcp_options_received *opt); 502f1673381SFlorian Westphal bool cookie_ecn_ok(const struct tcp_options_received *opt, 503f7b3bec6SFlorian Westphal const struct net *net, const struct dst_entry *dst); 5044dfc2817SFlorian Westphal 505c6aefafbSGlenn Griffin /* From net/ipv6/syncookies.c */ 5065c9f3023SJoe Perches int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, 50781eb6a14SPatrick McHardy u32 cookie); 5085c9f3023SJoe Perches struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); 509f1673381SFlorian Westphal 5105c9f3023SJoe Perches u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, 51181eb6a14SPatrick McHardy const struct tcphdr *th, u16 *mssp); 5125c9f3023SJoe Perches __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, 513c6aefafbSGlenn Griffin __u16 *mss); 514e05c82d3SEric Dumazet #endif 5151da177e4SLinus Torvalds /* tcp_output.c */ 5161da177e4SLinus Torvalds 5175c9f3023SJoe Perches void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, 5189e412ba7SIlpo Järvinen int nonagle); 5195c9f3023SJoe Perches bool tcp_may_send_now(struct sock *sk); 5205c9f3023SJoe Perches int __tcp_retransmit_skb(struct sock *, struct sk_buff *); 5215c9f3023SJoe Perches int tcp_retransmit_skb(struct sock *, struct sk_buff *); 5225c9f3023SJoe Perches void tcp_retransmit_timer(struct sock *sk); 5235c9f3023SJoe Perches void tcp_xmit_retransmit_queue(struct sock *); 5245c9f3023SJoe Perches void tcp_simple_retransmit(struct sock *); 5255c9f3023SJoe Perches int tcp_trim_head(struct sock *, struct sk_buff *, u32); 5266cc55e09SOctavian Purdila int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); 5271da177e4SLinus Torvalds 5285c9f3023SJoe Perches void tcp_send_probe0(struct sock *); 5295c9f3023SJoe Perches void tcp_send_partial(struct sock *); 5305c9f3023SJoe Perches int tcp_write_wakeup(struct sock *); 5315c9f3023SJoe Perches void tcp_send_fin(struct sock *sk); 5325c9f3023SJoe Perches void tcp_send_active_reset(struct sock *sk, gfp_t priority); 5335c9f3023SJoe Perches int tcp_send_synack(struct sock *); 5345c9f3023SJoe Perches void tcp_push_one(struct sock *, unsigned int mss_now); 5355c9f3023SJoe Perches void tcp_send_ack(struct sock *sk); 5365c9f3023SJoe Perches void tcp_send_delayed_ack(struct sock *sk); 5375c9f3023SJoe Perches void tcp_send_loss_probe(struct sock *sk); 5385c9f3023SJoe Perches bool tcp_schedule_loss_probe(struct sock *sk); 5391da177e4SLinus Torvalds 540a762a980SDavid S. Miller /* tcp_input.c */ 5415c9f3023SJoe Perches void tcp_resume_early_retransmit(struct sock *sk); 5425c9f3023SJoe Perches void tcp_rearm_rto(struct sock *sk); 5435c9f3023SJoe Perches void tcp_reset(struct sock *sk); 544a762a980SDavid S. Miller 5451da177e4SLinus Torvalds /* tcp_timer.c */ 5465c9f3023SJoe Perches void tcp_init_xmit_timers(struct sock *); 547463c84b9SArnaldo Carvalho de Melo static inline void tcp_clear_xmit_timers(struct sock *sk) 548463c84b9SArnaldo Carvalho de Melo { 549463c84b9SArnaldo Carvalho de Melo inet_csk_clear_xmit_timers(sk); 550463c84b9SArnaldo Carvalho de Melo } 5511da177e4SLinus Torvalds 5525c9f3023SJoe Perches unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); 5535c9f3023SJoe Perches unsigned int tcp_current_mss(struct sock *sk); 5540c54b85fSIlpo Järvinen 5550c54b85fSIlpo Järvinen /* Bound MSS / TSO packet size with the half of the window */ 5560c54b85fSIlpo Järvinen static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) 5570c54b85fSIlpo Järvinen { 55801f83d69SAlexey Kuznetsov int cutoff; 55901f83d69SAlexey Kuznetsov 56001f83d69SAlexey Kuznetsov /* When peer uses tiny windows, there is no use in packetizing 56101f83d69SAlexey Kuznetsov * to sub-MSS pieces for the sake of SWS or making sure there 56201f83d69SAlexey Kuznetsov * are enough packets in the pipe for fast recovery. 56301f83d69SAlexey Kuznetsov * 56401f83d69SAlexey Kuznetsov * On the other hand, for extremely large MSS devices, handling 56501f83d69SAlexey Kuznetsov * smaller than MSS windows in this way does make sense. 56601f83d69SAlexey Kuznetsov */ 56701f83d69SAlexey Kuznetsov if (tp->max_window >= 512) 56801f83d69SAlexey Kuznetsov cutoff = (tp->max_window >> 1); 56901f83d69SAlexey Kuznetsov else 57001f83d69SAlexey Kuznetsov cutoff = tp->max_window; 57101f83d69SAlexey Kuznetsov 57201f83d69SAlexey Kuznetsov if (cutoff && pktsize > cutoff) 57301f83d69SAlexey Kuznetsov return max_t(int, cutoff, 68U - tp->tcp_header_len); 5740c54b85fSIlpo Järvinen else 5750c54b85fSIlpo Järvinen return pktsize; 5760c54b85fSIlpo Järvinen } 5771da177e4SLinus Torvalds 57817b085eaSArnaldo Carvalho de Melo /* tcp.c */ 5795c9f3023SJoe Perches void tcp_get_info(const struct sock *, struct tcp_info *); 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds /* Read 'sendfile()'-style from a TCP socket */ 5821da177e4SLinus Torvalds typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, 5831da177e4SLinus Torvalds unsigned int, size_t); 5845c9f3023SJoe Perches int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, 5851da177e4SLinus Torvalds sk_read_actor_t recv_actor); 5861da177e4SLinus Torvalds 5875c9f3023SJoe Perches void tcp_initialize_rcv_mss(struct sock *sk); 5881da177e4SLinus Torvalds 5895c9f3023SJoe Perches int tcp_mtu_to_mss(struct sock *sk, int pmtu); 5905c9f3023SJoe Perches int tcp_mss_to_mtu(struct sock *sk, int mss); 5915c9f3023SJoe Perches void tcp_mtup_init(struct sock *sk); 5925c9f3023SJoe Perches void tcp_init_buffer_space(struct sock *sk); 5935d424d5aSJohn Heffner 594f1ecd5d9SDamian Lukowski static inline void tcp_bound_rto(const struct sock *sk) 595f1ecd5d9SDamian Lukowski { 596f1ecd5d9SDamian Lukowski if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) 597f1ecd5d9SDamian Lukowski inet_csk(sk)->icsk_rto = TCP_RTO_MAX; 598f1ecd5d9SDamian Lukowski } 599f1ecd5d9SDamian Lukowski 600f1ecd5d9SDamian Lukowski static inline u32 __tcp_set_rto(const struct tcp_sock *tp) 601f1ecd5d9SDamian Lukowski { 602740b0f18SEric Dumazet return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); 603f1ecd5d9SDamian Lukowski } 604f1ecd5d9SDamian Lukowski 60540efc6faSStephen Hemminger static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) 6061da177e4SLinus Torvalds { 6071da177e4SLinus Torvalds tp->pred_flags = htonl((tp->tcp_header_len << 26) | 6081da177e4SLinus Torvalds ntohl(TCP_FLAG_ACK) | 6091da177e4SLinus Torvalds snd_wnd); 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds 61240efc6faSStephen Hemminger static inline void tcp_fast_path_on(struct tcp_sock *tp) 6131da177e4SLinus Torvalds { 6141da177e4SLinus Torvalds __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); 6151da177e4SLinus Torvalds } 6161da177e4SLinus Torvalds 6179e412ba7SIlpo Järvinen static inline void tcp_fast_path_check(struct sock *sk) 6181da177e4SLinus Torvalds { 6199e412ba7SIlpo Järvinen struct tcp_sock *tp = tcp_sk(sk); 6209e412ba7SIlpo Järvinen 621b03efcfbSDavid S. Miller if (skb_queue_empty(&tp->out_of_order_queue) && 6221da177e4SLinus Torvalds tp->rcv_wnd && 6231da177e4SLinus Torvalds atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && 6241da177e4SLinus Torvalds !tp->urg_data) 6251da177e4SLinus Torvalds tcp_fast_path_on(tp); 6261da177e4SLinus Torvalds } 6271da177e4SLinus Torvalds 6280c266898SSatoru SATOH /* Compute the actual rto_min value */ 6290c266898SSatoru SATOH static inline u32 tcp_rto_min(struct sock *sk) 6300c266898SSatoru SATOH { 631cf533ea5SEric Dumazet const struct dst_entry *dst = __sk_dst_get(sk); 6320c266898SSatoru SATOH u32 rto_min = TCP_RTO_MIN; 6330c266898SSatoru SATOH 6340c266898SSatoru SATOH if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) 6350c266898SSatoru SATOH rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); 6360c266898SSatoru SATOH return rto_min; 6370c266898SSatoru SATOH } 6380c266898SSatoru SATOH 639740b0f18SEric Dumazet static inline u32 tcp_rto_min_us(struct sock *sk) 640740b0f18SEric Dumazet { 641740b0f18SEric Dumazet return jiffies_to_usecs(tcp_rto_min(sk)); 642740b0f18SEric Dumazet } 643740b0f18SEric Dumazet 64481164413SDaniel Borkmann static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) 64581164413SDaniel Borkmann { 64681164413SDaniel Borkmann return dst_metric_locked(dst, RTAX_CC_ALGO); 64781164413SDaniel Borkmann } 64881164413SDaniel Borkmann 6491da177e4SLinus Torvalds /* Compute the actual receive window we are currently advertising. 6501da177e4SLinus Torvalds * Rcv_nxt can be after the window if our peer push more data 6511da177e4SLinus Torvalds * than the offered window. 6521da177e4SLinus Torvalds */ 65340efc6faSStephen Hemminger static inline u32 tcp_receive_window(const struct tcp_sock *tp) 6541da177e4SLinus Torvalds { 6551da177e4SLinus Torvalds s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; 6561da177e4SLinus Torvalds 6571da177e4SLinus Torvalds if (win < 0) 6581da177e4SLinus Torvalds win = 0; 6591da177e4SLinus Torvalds return (u32) win; 6601da177e4SLinus Torvalds } 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds /* Choose a new window, without checks for shrinking, and without 6631da177e4SLinus Torvalds * scaling applied to the result. The caller does these things 6641da177e4SLinus Torvalds * if necessary. This is a "raw" window selection. 6651da177e4SLinus Torvalds */ 6665c9f3023SJoe Perches u32 __tcp_select_window(struct sock *sk); 6671da177e4SLinus Torvalds 668ee995283SPavel Emelyanov void tcp_send_window_probe(struct sock *sk); 669ee995283SPavel Emelyanov 6701da177e4SLinus Torvalds /* TCP timestamps are only 32-bits, this causes a slight 6711da177e4SLinus Torvalds * complication on 64-bit systems since we store a snapshot 67231f34269SStephen Hemminger * of jiffies in the buffer control blocks below. We decided 67331f34269SStephen Hemminger * to use only the low 32-bits of jiffies and hide the ugly 6741da177e4SLinus Torvalds * casts with the following macro. 6751da177e4SLinus Torvalds */ 6761da177e4SLinus Torvalds #define tcp_time_stamp ((__u32)(jiffies)) 6771da177e4SLinus Torvalds 6787faee5c0SEric Dumazet static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) 6797faee5c0SEric Dumazet { 6807faee5c0SEric Dumazet return skb->skb_mstamp.stamp_jiffies; 6817faee5c0SEric Dumazet } 6827faee5c0SEric Dumazet 6837faee5c0SEric Dumazet 684a3433f35SChangli Gao #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) 685a3433f35SChangli Gao 686a3433f35SChangli Gao #define TCPHDR_FIN 0x01 687a3433f35SChangli Gao #define TCPHDR_SYN 0x02 688a3433f35SChangli Gao #define TCPHDR_RST 0x04 689a3433f35SChangli Gao #define TCPHDR_PSH 0x08 690a3433f35SChangli Gao #define TCPHDR_ACK 0x10 691a3433f35SChangli Gao #define TCPHDR_URG 0x20 692a3433f35SChangli Gao #define TCPHDR_ECE 0x40 693a3433f35SChangli Gao #define TCPHDR_CWR 0x80 694a3433f35SChangli Gao 695caa20d9aSStephen Hemminger /* This is what the send packet queuing engine uses to pass 696f86586faSEric Dumazet * TCP per-packet control information to the transmission code. 697f86586faSEric Dumazet * We also store the host-order sequence numbers in here too. 698f86586faSEric Dumazet * This is 44 bytes if IPV6 is enabled. 699f86586faSEric Dumazet * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. 7001da177e4SLinus Torvalds */ 7011da177e4SLinus Torvalds struct tcp_skb_cb { 7021da177e4SLinus Torvalds __u32 seq; /* Starting sequence number */ 7031da177e4SLinus Torvalds __u32 end_seq; /* SEQ + FIN + SYN + datalen */ 704cd7d8498SEric Dumazet union { 705cd7d8498SEric Dumazet /* Note : tcp_tw_isn is used in input path only 706cd7d8498SEric Dumazet * (isn chosen by tcp_timewait_state_process()) 707cd7d8498SEric Dumazet * 708cd7d8498SEric Dumazet * tcp_gso_segs is used in write queue only, 709cd7d8498SEric Dumazet * cf tcp_skb_pcount() 710cd7d8498SEric Dumazet */ 711cd7d8498SEric Dumazet __u32 tcp_tw_isn; 712cd7d8498SEric Dumazet __u32 tcp_gso_segs; 713cd7d8498SEric Dumazet }; 7144de075e0SEric Dumazet __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ 715f4f9f6e7SNeal Cardwell 7161da177e4SLinus Torvalds __u8 sacked; /* State flags for SACK/FACK. */ 7171da177e4SLinus Torvalds #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ 7181da177e4SLinus Torvalds #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ 7191da177e4SLinus Torvalds #define TCPCB_LOST 0x04 /* SKB is lost */ 7201da177e4SLinus Torvalds #define TCPCB_TAGBITS 0x07 /* All tag bits */ 7219d186cacSAndrey Vagin #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ 7221da177e4SLinus Torvalds #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ 7239d186cacSAndrey Vagin #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ 7249d186cacSAndrey Vagin TCPCB_REPAIRED) 7251da177e4SLinus Torvalds 726f4f9f6e7SNeal Cardwell __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ 727f4f9f6e7SNeal Cardwell /* 1 byte hole */ 7281da177e4SLinus Torvalds __u32 ack_seq; /* Sequence number ACK'd */ 729971f10ecSEric Dumazet union { 730971f10ecSEric Dumazet struct inet_skb_parm h4; 731971f10ecSEric Dumazet #if IS_ENABLED(CONFIG_IPV6) 732971f10ecSEric Dumazet struct inet6_skb_parm h6; 733971f10ecSEric Dumazet #endif 734971f10ecSEric Dumazet } header; /* For incoming frames */ 7351da177e4SLinus Torvalds }; 7361da177e4SLinus Torvalds 7371da177e4SLinus Torvalds #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) 7381da177e4SLinus Torvalds 739870c3151SEric Dumazet 740815afe17SEric Dumazet #if IS_ENABLED(CONFIG_IPV6) 741870c3151SEric Dumazet /* This is the variant of inet6_iif() that must be used by TCP, 742870c3151SEric Dumazet * as TCP moves IP6CB into a different location in skb->cb[] 743870c3151SEric Dumazet */ 744870c3151SEric Dumazet static inline int tcp_v6_iif(const struct sk_buff *skb) 745870c3151SEric Dumazet { 746870c3151SEric Dumazet return TCP_SKB_CB(skb)->header.h6.iif; 747870c3151SEric Dumazet } 748815afe17SEric Dumazet #endif 749870c3151SEric Dumazet 7501da177e4SLinus Torvalds /* Due to TSO, an SKB can be composed of multiple actual 7511da177e4SLinus Torvalds * packets. To keep these tracked properly, we use this. 7521da177e4SLinus Torvalds */ 7531da177e4SLinus Torvalds static inline int tcp_skb_pcount(const struct sk_buff *skb) 7541da177e4SLinus Torvalds { 755cd7d8498SEric Dumazet return TCP_SKB_CB(skb)->tcp_gso_segs; 756cd7d8498SEric Dumazet } 757cd7d8498SEric Dumazet 758cd7d8498SEric Dumazet static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) 759cd7d8498SEric Dumazet { 760cd7d8498SEric Dumazet TCP_SKB_CB(skb)->tcp_gso_segs = segs; 761cd7d8498SEric Dumazet } 762cd7d8498SEric Dumazet 763cd7d8498SEric Dumazet static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) 764cd7d8498SEric Dumazet { 765cd7d8498SEric Dumazet TCP_SKB_CB(skb)->tcp_gso_segs += segs; 7661da177e4SLinus Torvalds } 7671da177e4SLinus Torvalds 7681da177e4SLinus Torvalds /* This is valid iff tcp_skb_pcount() > 1. */ 7691da177e4SLinus Torvalds static inline int tcp_skb_mss(const struct sk_buff *skb) 7701da177e4SLinus Torvalds { 7717967168cSHerbert Xu return skb_shinfo(skb)->gso_size; 7721da177e4SLinus Torvalds } 7731da177e4SLinus Torvalds 774317a76f9SStephen Hemminger /* Events passed to congestion control interface */ 775317a76f9SStephen Hemminger enum tcp_ca_event { 776317a76f9SStephen Hemminger CA_EVENT_TX_START, /* first transmit when no packets in flight */ 777317a76f9SStephen Hemminger CA_EVENT_CWND_RESTART, /* congestion window restart */ 778317a76f9SStephen Hemminger CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ 779317a76f9SStephen Hemminger CA_EVENT_LOSS, /* loss timeout */ 7809890092eSFlorian Westphal CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ 7819890092eSFlorian Westphal CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ 7829890092eSFlorian Westphal CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ 7839890092eSFlorian Westphal CA_EVENT_NON_DELAYED_ACK, 7847354c8c3SFlorian Westphal }; 7857354c8c3SFlorian Westphal 7869890092eSFlorian Westphal /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ 7877354c8c3SFlorian Westphal enum tcp_ca_ack_event_flags { 7889890092eSFlorian Westphal CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ 7899890092eSFlorian Westphal CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ 7909890092eSFlorian Westphal CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ 791317a76f9SStephen Hemminger }; 792317a76f9SStephen Hemminger 793317a76f9SStephen Hemminger /* 794317a76f9SStephen Hemminger * Interface for adding new TCP congestion control handlers 795317a76f9SStephen Hemminger */ 796317a76f9SStephen Hemminger #define TCP_CA_NAME_MAX 16 7973ff825b2SStephen Hemminger #define TCP_CA_MAX 128 7983ff825b2SStephen Hemminger #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) 7993ff825b2SStephen Hemminger 800c5c6a8abSDaniel Borkmann #define TCP_CA_UNSPEC 0 801c5c6a8abSDaniel Borkmann 80230e502a3SDaniel Borkmann /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ 803164891aaSStephen Hemminger #define TCP_CONG_NON_RESTRICTED 0x1 80430e502a3SDaniel Borkmann /* Requires ECN/ECT set on all packets */ 80530e502a3SDaniel Borkmann #define TCP_CONG_NEEDS_ECN 0x2 806164891aaSStephen Hemminger 807317a76f9SStephen Hemminger struct tcp_congestion_ops { 808317a76f9SStephen Hemminger struct list_head list; 809c5c6a8abSDaniel Borkmann u32 key; 810c5c6a8abSDaniel Borkmann u32 flags; 811317a76f9SStephen Hemminger 812317a76f9SStephen Hemminger /* initialize private data (optional) */ 8136687e988SArnaldo Carvalho de Melo void (*init)(struct sock *sk); 814317a76f9SStephen Hemminger /* cleanup private data (optional) */ 8156687e988SArnaldo Carvalho de Melo void (*release)(struct sock *sk); 816317a76f9SStephen Hemminger 817317a76f9SStephen Hemminger /* return slow start threshold (required) */ 8186687e988SArnaldo Carvalho de Melo u32 (*ssthresh)(struct sock *sk); 819317a76f9SStephen Hemminger /* do new cwnd calculation (required) */ 82024901551SEric Dumazet void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); 821317a76f9SStephen Hemminger /* call before changing ca_state (optional) */ 8226687e988SArnaldo Carvalho de Melo void (*set_state)(struct sock *sk, u8 new_state); 823317a76f9SStephen Hemminger /* call when cwnd event occurs (optional) */ 8246687e988SArnaldo Carvalho de Melo void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); 8257354c8c3SFlorian Westphal /* call when ack arrives (optional) */ 8267354c8c3SFlorian Westphal void (*in_ack_event)(struct sock *sk, u32 flags); 827317a76f9SStephen Hemminger /* new value of cwnd after loss (optional) */ 8286687e988SArnaldo Carvalho de Melo u32 (*undo_cwnd)(struct sock *sk); 829317a76f9SStephen Hemminger /* hook for packet ack accounting (optional) */ 83030cfd0baSStephen Hemminger void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us); 83173c1f4a0SArnaldo Carvalho de Melo /* get info for inet_diag (optional) */ 8326687e988SArnaldo Carvalho de Melo void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb); 833317a76f9SStephen Hemminger 834317a76f9SStephen Hemminger char name[TCP_CA_NAME_MAX]; 835317a76f9SStephen Hemminger struct module *owner; 836317a76f9SStephen Hemminger }; 837317a76f9SStephen Hemminger 8385c9f3023SJoe Perches int tcp_register_congestion_control(struct tcp_congestion_ops *type); 8395c9f3023SJoe Perches void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); 840317a76f9SStephen Hemminger 84155d8694fSFlorian Westphal void tcp_assign_congestion_control(struct sock *sk); 8425c9f3023SJoe Perches void tcp_init_congestion_control(struct sock *sk); 8435c9f3023SJoe Perches void tcp_cleanup_congestion_control(struct sock *sk); 8445c9f3023SJoe Perches int tcp_set_default_congestion_control(const char *name); 8455c9f3023SJoe Perches void tcp_get_default_congestion_control(char *name); 8465c9f3023SJoe Perches void tcp_get_available_congestion_control(char *buf, size_t len); 8475c9f3023SJoe Perches void tcp_get_allowed_congestion_control(char *buf, size_t len); 8485c9f3023SJoe Perches int tcp_set_allowed_congestion_control(char *allowed); 8495c9f3023SJoe Perches int tcp_set_congestion_control(struct sock *sk, const char *name); 850e73ebb08SNeal Cardwell u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); 851e73ebb08SNeal Cardwell void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); 852317a76f9SStephen Hemminger 8535c9f3023SJoe Perches u32 tcp_reno_ssthresh(struct sock *sk); 85424901551SEric Dumazet void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); 855a8acfbacSDavid S. Miller extern struct tcp_congestion_ops tcp_reno; 856317a76f9SStephen Hemminger 857c5c6a8abSDaniel Borkmann struct tcp_congestion_ops *tcp_ca_find_key(u32 key); 858c5c6a8abSDaniel Borkmann u32 tcp_ca_get_key_by_name(const char *name); 859ea697639SDaniel Borkmann #ifdef CONFIG_INET 860c5c6a8abSDaniel Borkmann char *tcp_ca_get_name_by_key(u32 key, char *buffer); 861ea697639SDaniel Borkmann #else 862ea697639SDaniel Borkmann static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) 863ea697639SDaniel Borkmann { 864ea697639SDaniel Borkmann return NULL; 865ea697639SDaniel Borkmann } 866ea697639SDaniel Borkmann #endif 867c5c6a8abSDaniel Borkmann 86830e502a3SDaniel Borkmann static inline bool tcp_ca_needs_ecn(const struct sock *sk) 86930e502a3SDaniel Borkmann { 87030e502a3SDaniel Borkmann const struct inet_connection_sock *icsk = inet_csk(sk); 87130e502a3SDaniel Borkmann 87230e502a3SDaniel Borkmann return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; 87330e502a3SDaniel Borkmann } 87430e502a3SDaniel Borkmann 8756687e988SArnaldo Carvalho de Melo static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) 876317a76f9SStephen Hemminger { 8776687e988SArnaldo Carvalho de Melo struct inet_connection_sock *icsk = inet_csk(sk); 8786687e988SArnaldo Carvalho de Melo 8796687e988SArnaldo Carvalho de Melo if (icsk->icsk_ca_ops->set_state) 8806687e988SArnaldo Carvalho de Melo icsk->icsk_ca_ops->set_state(sk, ca_state); 8816687e988SArnaldo Carvalho de Melo icsk->icsk_ca_state = ca_state; 882317a76f9SStephen Hemminger } 883317a76f9SStephen Hemminger 8846687e988SArnaldo Carvalho de Melo static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) 885317a76f9SStephen Hemminger { 8866687e988SArnaldo Carvalho de Melo const struct inet_connection_sock *icsk = inet_csk(sk); 8876687e988SArnaldo Carvalho de Melo 8886687e988SArnaldo Carvalho de Melo if (icsk->icsk_ca_ops->cwnd_event) 8896687e988SArnaldo Carvalho de Melo icsk->icsk_ca_ops->cwnd_event(sk, event); 890317a76f9SStephen Hemminger } 891317a76f9SStephen Hemminger 892e60402d0SIlpo Järvinen /* These functions determine how the current flow behaves in respect of SACK 893e60402d0SIlpo Järvinen * handling. SACK is negotiated with the peer, and therefore it can vary 894e60402d0SIlpo Järvinen * between different flows. 895e60402d0SIlpo Järvinen * 896e60402d0SIlpo Järvinen * tcp_is_sack - SACK enabled 897e60402d0SIlpo Järvinen * tcp_is_reno - No SACK 898e60402d0SIlpo Järvinen * tcp_is_fack - FACK enabled, implies SACK enabled 899e60402d0SIlpo Järvinen */ 900e60402d0SIlpo Järvinen static inline int tcp_is_sack(const struct tcp_sock *tp) 901e60402d0SIlpo Järvinen { 902e60402d0SIlpo Järvinen return tp->rx_opt.sack_ok; 903e60402d0SIlpo Järvinen } 904e60402d0SIlpo Järvinen 905a2a385d6SEric Dumazet static inline bool tcp_is_reno(const struct tcp_sock *tp) 906e60402d0SIlpo Järvinen { 907e60402d0SIlpo Järvinen return !tcp_is_sack(tp); 908e60402d0SIlpo Järvinen } 909e60402d0SIlpo Järvinen 910a2a385d6SEric Dumazet static inline bool tcp_is_fack(const struct tcp_sock *tp) 911e60402d0SIlpo Järvinen { 912ab56222aSVijay Subramanian return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; 913e60402d0SIlpo Järvinen } 914e60402d0SIlpo Järvinen 915e60402d0SIlpo Järvinen static inline void tcp_enable_fack(struct tcp_sock *tp) 916e60402d0SIlpo Järvinen { 917ab56222aSVijay Subramanian tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; 918e60402d0SIlpo Järvinen } 919e60402d0SIlpo Järvinen 920eed530b6SYuchung Cheng /* TCP early-retransmit (ER) is similar to but more conservative than 921eed530b6SYuchung Cheng * the thin-dupack feature. Enable ER only if thin-dupack is disabled. 922eed530b6SYuchung Cheng */ 923eed530b6SYuchung Cheng static inline void tcp_enable_early_retrans(struct tcp_sock *tp) 924eed530b6SYuchung Cheng { 925eed530b6SYuchung Cheng tp->do_early_retrans = sysctl_tcp_early_retrans && 9266ba8a3b1SNandita Dukkipati sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && 9276ba8a3b1SNandita Dukkipati sysctl_tcp_reordering == 3; 928eed530b6SYuchung Cheng } 929eed530b6SYuchung Cheng 930eed530b6SYuchung Cheng static inline void tcp_disable_early_retrans(struct tcp_sock *tp) 931eed530b6SYuchung Cheng { 932eed530b6SYuchung Cheng tp->do_early_retrans = 0; 933eed530b6SYuchung Cheng } 934eed530b6SYuchung Cheng 93583ae4088SIlpo Järvinen static inline unsigned int tcp_left_out(const struct tcp_sock *tp) 93683ae4088SIlpo Järvinen { 93783ae4088SIlpo Järvinen return tp->sacked_out + tp->lost_out; 93883ae4088SIlpo Järvinen } 93983ae4088SIlpo Järvinen 9401da177e4SLinus Torvalds /* This determines how many packets are "in the network" to the best 9411da177e4SLinus Torvalds * of our knowledge. In many cases it is conservative, but where 9421da177e4SLinus Torvalds * detailed information is available from the receiver (via SACK 9431da177e4SLinus Torvalds * blocks etc.) we can make more aggressive calculations. 9441da177e4SLinus Torvalds * 9451da177e4SLinus Torvalds * Use this for decisions involving congestion control, use just 9461da177e4SLinus Torvalds * tp->packets_out to determine if the send queue is empty or not. 9471da177e4SLinus Torvalds * 9481da177e4SLinus Torvalds * Read this equation as: 9491da177e4SLinus Torvalds * 9501da177e4SLinus Torvalds * "Packets sent once on transmission queue" MINUS 9511da177e4SLinus Torvalds * "Packets left network, but not honestly ACKed yet" PLUS 9521da177e4SLinus Torvalds * "Packets fast retransmitted" 9531da177e4SLinus Torvalds */ 95440efc6faSStephen Hemminger static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) 9551da177e4SLinus Torvalds { 95683ae4088SIlpo Järvinen return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; 9571da177e4SLinus Torvalds } 9581da177e4SLinus Torvalds 9590b6a05c1SIlpo Järvinen #define TCP_INFINITE_SSTHRESH 0x7fffffff 9600b6a05c1SIlpo Järvinen 9610b6a05c1SIlpo Järvinen static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) 9620b6a05c1SIlpo Järvinen { 9630b6a05c1SIlpo Järvinen return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; 9640b6a05c1SIlpo Järvinen } 9650b6a05c1SIlpo Järvinen 966684bad11SYuchung Cheng static inline bool tcp_in_cwnd_reduction(const struct sock *sk) 967684bad11SYuchung Cheng { 968684bad11SYuchung Cheng return (TCPF_CA_CWR | TCPF_CA_Recovery) & 969684bad11SYuchung Cheng (1 << inet_csk(sk)->icsk_ca_state); 970684bad11SYuchung Cheng } 971684bad11SYuchung Cheng 9721da177e4SLinus Torvalds /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. 973684bad11SYuchung Cheng * The exception is cwnd reduction phase, when cwnd is decreasing towards 9741da177e4SLinus Torvalds * ssthresh. 9751da177e4SLinus Torvalds */ 9766687e988SArnaldo Carvalho de Melo static inline __u32 tcp_current_ssthresh(const struct sock *sk) 9771da177e4SLinus Torvalds { 9786687e988SArnaldo Carvalho de Melo const struct tcp_sock *tp = tcp_sk(sk); 979cf533ea5SEric Dumazet 980684bad11SYuchung Cheng if (tcp_in_cwnd_reduction(sk)) 9811da177e4SLinus Torvalds return tp->snd_ssthresh; 9821da177e4SLinus Torvalds else 9831da177e4SLinus Torvalds return max(tp->snd_ssthresh, 9841da177e4SLinus Torvalds ((tp->snd_cwnd >> 1) + 9851da177e4SLinus Torvalds (tp->snd_cwnd >> 2))); 9861da177e4SLinus Torvalds } 9871da177e4SLinus Torvalds 988b9c4595bSIlpo Järvinen /* Use define here intentionally to get WARN_ON location shown at the caller */ 989b9c4595bSIlpo Järvinen #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) 9901da177e4SLinus Torvalds 9915ee2c941SChristoph Paasch void tcp_enter_cwr(struct sock *sk); 9925c9f3023SJoe Perches __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); 9931da177e4SLinus Torvalds 9946b5a5c0dSNeal Cardwell /* The maximum number of MSS of available cwnd for which TSO defers 9956b5a5c0dSNeal Cardwell * sending if not using sysctl_tcp_tso_win_divisor. 9966b5a5c0dSNeal Cardwell */ 9976b5a5c0dSNeal Cardwell static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) 9986b5a5c0dSNeal Cardwell { 9996b5a5c0dSNeal Cardwell return 3; 10006b5a5c0dSNeal Cardwell } 10016b5a5c0dSNeal Cardwell 10021da177e4SLinus Torvalds /* Slow start with delack produces 3 packets of burst, so that 1003dd9e0ddaSJohn Heffner * it is safe "de facto". This will be the default - same as 1004dd9e0ddaSJohn Heffner * the default reordering threshold - but if reordering increases, 1005dd9e0ddaSJohn Heffner * we must be able to allow cwnd to burst at least this much in order 1006dd9e0ddaSJohn Heffner * to not pull it back when holes are filled. 10071da177e4SLinus Torvalds */ 10081da177e4SLinus Torvalds static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) 10091da177e4SLinus Torvalds { 1010dd9e0ddaSJohn Heffner return tp->reordering; 10111da177e4SLinus Torvalds } 10121da177e4SLinus Torvalds 101390840defSIlpo Järvinen /* Returns end sequence number of the receiver's advertised window */ 101490840defSIlpo Järvinen static inline u32 tcp_wnd_end(const struct tcp_sock *tp) 101590840defSIlpo Järvinen { 101690840defSIlpo Järvinen return tp->snd_una + tp->snd_wnd; 101790840defSIlpo Järvinen } 1018e114a710SEric Dumazet 1019e114a710SEric Dumazet /* We follow the spirit of RFC2861 to validate cwnd but implement a more 1020e114a710SEric Dumazet * flexible approach. The RFC suggests cwnd should not be raised unless 1021ca8a2263SNeal Cardwell * it was fully used previously. And that's exactly what we do in 1022ca8a2263SNeal Cardwell * congestion avoidance mode. But in slow start we allow cwnd to grow 1023ca8a2263SNeal Cardwell * as long as the application has used half the cwnd. 1024e114a710SEric Dumazet * Example : 1025e114a710SEric Dumazet * cwnd is 10 (IW10), but application sends 9 frames. 1026e114a710SEric Dumazet * We allow cwnd to reach 18 when all frames are ACKed. 1027e114a710SEric Dumazet * This check is safe because it's as aggressive as slow start which already 1028e114a710SEric Dumazet * risks 100% overshoot. The advantage is that we discourage application to 1029e114a710SEric Dumazet * either send more filler packets or data to artificially blow up the cwnd 1030e114a710SEric Dumazet * usage, and allow application-limited process to probe bw more aggressively. 1031e114a710SEric Dumazet */ 103224901551SEric Dumazet static inline bool tcp_is_cwnd_limited(const struct sock *sk) 1033e114a710SEric Dumazet { 1034e114a710SEric Dumazet const struct tcp_sock *tp = tcp_sk(sk); 1035e114a710SEric Dumazet 1036ca8a2263SNeal Cardwell /* If in slow start, ensure cwnd grows to twice what was ACKed. */ 1037ca8a2263SNeal Cardwell if (tp->snd_cwnd <= tp->snd_ssthresh) 1038ca8a2263SNeal Cardwell return tp->snd_cwnd < 2 * tp->max_packets_out; 1039ca8a2263SNeal Cardwell 1040ca8a2263SNeal Cardwell return tp->is_cwnd_limited; 1041e114a710SEric Dumazet } 1042f4805edeSStephen Hemminger 10439e412ba7SIlpo Järvinen static inline void tcp_check_probe_timer(struct sock *sk) 10441da177e4SLinus Torvalds { 1045cf533ea5SEric Dumazet const struct tcp_sock *tp = tcp_sk(sk); 1046463c84b9SArnaldo Carvalho de Melo const struct inet_connection_sock *icsk = inet_csk(sk); 10479e412ba7SIlpo Järvinen 1048463c84b9SArnaldo Carvalho de Melo if (!tp->packets_out && !icsk->icsk_pending) 10493f421baaSArnaldo Carvalho de Melo inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, 10503f421baaSArnaldo Carvalho de Melo icsk->icsk_rto, TCP_RTO_MAX); 10511da177e4SLinus Torvalds } 10521da177e4SLinus Torvalds 1053ee7537b6SHantzis Fotis static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) 10541da177e4SLinus Torvalds { 10551da177e4SLinus Torvalds tp->snd_wl1 = seq; 10561da177e4SLinus Torvalds } 10571da177e4SLinus Torvalds 1058ee7537b6SHantzis Fotis static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) 10591da177e4SLinus Torvalds { 10601da177e4SLinus Torvalds tp->snd_wl1 = seq; 10611da177e4SLinus Torvalds } 10621da177e4SLinus Torvalds 10631da177e4SLinus Torvalds /* 10641da177e4SLinus Torvalds * Calculate(/check) TCP checksum 10651da177e4SLinus Torvalds */ 1066ba7808eaSFrederik Deweerdt static inline __sum16 tcp_v4_check(int len, __be32 saddr, 1067ba7808eaSFrederik Deweerdt __be32 daddr, __wsum base) 10681da177e4SLinus Torvalds { 10691da177e4SLinus Torvalds return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 1072b51655b9SAl Viro static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) 10731da177e4SLinus Torvalds { 1074fb286bb2SHerbert Xu return __skb_checksum_complete(skb); 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds 1077a2a385d6SEric Dumazet static inline bool tcp_checksum_complete(struct sk_buff *skb) 10781da177e4SLinus Torvalds { 107960476372SHerbert Xu return !skb_csum_unnecessary(skb) && 10801da177e4SLinus Torvalds __tcp_checksum_complete(skb); 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds 10831da177e4SLinus Torvalds /* Prequeue for VJ style copy to user, combined with checksumming. */ 10841da177e4SLinus Torvalds 108540efc6faSStephen Hemminger static inline void tcp_prequeue_init(struct tcp_sock *tp) 10861da177e4SLinus Torvalds { 10871da177e4SLinus Torvalds tp->ucopy.task = NULL; 10881da177e4SLinus Torvalds tp->ucopy.len = 0; 10891da177e4SLinus Torvalds tp->ucopy.memory = 0; 10901da177e4SLinus Torvalds skb_queue_head_init(&tp->ucopy.prequeue); 10911da177e4SLinus Torvalds } 10921da177e4SLinus Torvalds 10935c9f3023SJoe Perches bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); 10941da177e4SLinus Torvalds 10951da177e4SLinus Torvalds #undef STATE_TRACE 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds #ifdef STATE_TRACE 10981da177e4SLinus Torvalds static const char *statename[]={ 10991da177e4SLinus Torvalds "Unused","Established","Syn Sent","Syn Recv", 11001da177e4SLinus Torvalds "Fin Wait 1","Fin Wait 2","Time Wait", "Close", 11011da177e4SLinus Torvalds "Close Wait","Last ACK","Listen","Closing" 11021da177e4SLinus Torvalds }; 11031da177e4SLinus Torvalds #endif 11045c9f3023SJoe Perches void tcp_set_state(struct sock *sk, int state); 11051da177e4SLinus Torvalds 11065c9f3023SJoe Perches void tcp_done(struct sock *sk); 11071da177e4SLinus Torvalds 110840efc6faSStephen Hemminger static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) 11091da177e4SLinus Torvalds { 11101da177e4SLinus Torvalds rx_opt->dsack = 0; 11111da177e4SLinus Torvalds rx_opt->num_sacks = 0; 11121da177e4SLinus Torvalds } 11131da177e4SLinus Torvalds 11145c9f3023SJoe Perches u32 tcp_default_init_rwnd(u32 mss); 111585f16525SYuchung Cheng 11161da177e4SLinus Torvalds /* Determine a window scaling and initial window to offer. */ 11175c9f3023SJoe Perches void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, 11185c9f3023SJoe Perches __u32 *window_clamp, int wscale_ok, 11195c9f3023SJoe Perches __u8 *rcv_wscale, __u32 init_rcv_wnd); 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds static inline int tcp_win_from_space(int space) 11221da177e4SLinus Torvalds { 11231da177e4SLinus Torvalds return sysctl_tcp_adv_win_scale<=0 ? 11241da177e4SLinus Torvalds (space>>(-sysctl_tcp_adv_win_scale)) : 11251da177e4SLinus Torvalds space - (space>>sysctl_tcp_adv_win_scale); 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds /* Note: caller must be prepared to deal with negative returns */ 11291da177e4SLinus Torvalds static inline int tcp_space(const struct sock *sk) 11301da177e4SLinus Torvalds { 11311da177e4SLinus Torvalds return tcp_win_from_space(sk->sk_rcvbuf - 11321da177e4SLinus Torvalds atomic_read(&sk->sk_rmem_alloc)); 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds 11351da177e4SLinus Torvalds static inline int tcp_full_space(const struct sock *sk) 11361da177e4SLinus Torvalds { 11371da177e4SLinus Torvalds return tcp_win_from_space(sk->sk_rcvbuf); 11381da177e4SLinus Torvalds } 11391da177e4SLinus Torvalds 1140843f4a55SYuchung Cheng extern void tcp_openreq_init_rwin(struct request_sock *req, 1141843f4a55SYuchung Cheng struct sock *sk, struct dst_entry *dst); 1142843f4a55SYuchung Cheng 11435c9f3023SJoe Perches void tcp_enter_memory_pressure(struct sock *sk); 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds static inline int keepalive_intvl_when(const struct tcp_sock *tp) 11461da177e4SLinus Torvalds { 11471da177e4SLinus Torvalds return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; 11481da177e4SLinus Torvalds } 11491da177e4SLinus Torvalds 11501da177e4SLinus Torvalds static inline int keepalive_time_when(const struct tcp_sock *tp) 11511da177e4SLinus Torvalds { 11521da177e4SLinus Torvalds return tp->keepalive_time ? : sysctl_tcp_keepalive_time; 11531da177e4SLinus Torvalds } 11541da177e4SLinus Torvalds 1155df19a626SEric Dumazet static inline int keepalive_probes(const struct tcp_sock *tp) 1156df19a626SEric Dumazet { 1157df19a626SEric Dumazet return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes; 1158df19a626SEric Dumazet } 1159df19a626SEric Dumazet 11606c37e5deSFlavio Leitner static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) 11616c37e5deSFlavio Leitner { 11626c37e5deSFlavio Leitner const struct inet_connection_sock *icsk = &tp->inet_conn; 11636c37e5deSFlavio Leitner 11646c37e5deSFlavio Leitner return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, 11656c37e5deSFlavio Leitner tcp_time_stamp - tp->rcv_tstamp); 11666c37e5deSFlavio Leitner } 11676c37e5deSFlavio Leitner 1168463c84b9SArnaldo Carvalho de Melo static inline int tcp_fin_time(const struct sock *sk) 11691da177e4SLinus Torvalds { 1170463c84b9SArnaldo Carvalho de Melo int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout; 1171463c84b9SArnaldo Carvalho de Melo const int rto = inet_csk(sk)->icsk_rto; 11721da177e4SLinus Torvalds 1173463c84b9SArnaldo Carvalho de Melo if (fin_timeout < (rto << 2) - (rto >> 1)) 1174463c84b9SArnaldo Carvalho de Melo fin_timeout = (rto << 2) - (rto >> 1); 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds return fin_timeout; 11771da177e4SLinus Torvalds } 11781da177e4SLinus Torvalds 1179a2a385d6SEric Dumazet static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, 1180c887e6d2SIlpo Järvinen int paws_win) 11811da177e4SLinus Torvalds { 1182c887e6d2SIlpo Järvinen if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) 1183a2a385d6SEric Dumazet return true; 1184c887e6d2SIlpo Järvinen if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) 1185a2a385d6SEric Dumazet return true; 1186bc2ce894SEric Dumazet /* 1187bc2ce894SEric Dumazet * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, 1188bc2ce894SEric Dumazet * then following tcp messages have valid values. Ignore 0 value, 1189bc2ce894SEric Dumazet * or else 'negative' tsval might forbid us to accept their packets. 1190bc2ce894SEric Dumazet */ 1191bc2ce894SEric Dumazet if (!rx_opt->ts_recent) 1192a2a385d6SEric Dumazet return true; 1193a2a385d6SEric Dumazet return false; 1194c887e6d2SIlpo Järvinen } 1195c887e6d2SIlpo Järvinen 1196a2a385d6SEric Dumazet static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, 1197c887e6d2SIlpo Järvinen int rst) 1198c887e6d2SIlpo Järvinen { 1199c887e6d2SIlpo Järvinen if (tcp_paws_check(rx_opt, 0)) 1200a2a385d6SEric Dumazet return false; 12011da177e4SLinus Torvalds 12021da177e4SLinus Torvalds /* RST segments are not recommended to carry timestamp, 12031da177e4SLinus Torvalds and, if they do, it is recommended to ignore PAWS because 12041da177e4SLinus Torvalds "their cleanup function should take precedence over timestamps." 12051da177e4SLinus Torvalds Certainly, it is mistake. It is necessary to understand the reasons 12061da177e4SLinus Torvalds of this constraint to relax it: if peer reboots, clock may go 12071da177e4SLinus Torvalds out-of-sync and half-open connections will not be reset. 12081da177e4SLinus Torvalds Actually, the problem would be not existing if all 12091da177e4SLinus Torvalds the implementations followed draft about maintaining clock 12101da177e4SLinus Torvalds via reboots. Linux-2.2 DOES NOT! 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds However, we can relax time bounds for RST segments to MSL. 12131da177e4SLinus Torvalds */ 12149d729f72SJames Morris if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) 1215a2a385d6SEric Dumazet return false; 1216a2a385d6SEric Dumazet return true; 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds 12197970ddc8SEric Dumazet bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, 12207970ddc8SEric Dumazet int mib_idx, u32 *last_oow_ack_time); 1221032ee423SNeal Cardwell 1222a9c19329SPavel Emelyanov static inline void tcp_mib_init(struct net *net) 12231da177e4SLinus Torvalds { 12241da177e4SLinus Torvalds /* See RFC 2012 */ 1225cf1100a7SPavel Emelyanov TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1); 1226cf1100a7SPavel Emelyanov TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); 1227cf1100a7SPavel Emelyanov TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); 1228cf1100a7SPavel Emelyanov TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1); 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds 12316a438bbeSStephen Hemminger /* from STCP */ 1232ef9da47cSIlpo Järvinen static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) 12330800f170SDavid S. Miller { 12346a438bbeSStephen Hemminger tp->lost_skb_hint = NULL; 1235ef9da47cSIlpo Järvinen } 1236ef9da47cSIlpo Järvinen 1237ef9da47cSIlpo Järvinen static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) 1238ef9da47cSIlpo Järvinen { 1239ef9da47cSIlpo Järvinen tcp_clear_retrans_hints_partial(tp); 12406a438bbeSStephen Hemminger tp->retransmit_skb_hint = NULL; 1241b7689205SIlpo Järvinen } 1242b7689205SIlpo Järvinen 1243cfb6eeb4SYOSHIFUJI Hideaki /* MD5 Signature */ 1244cfb6eeb4SYOSHIFUJI Hideaki struct crypto_hash; 1245cfb6eeb4SYOSHIFUJI Hideaki 1246a915da9bSEric Dumazet union tcp_md5_addr { 1247a915da9bSEric Dumazet struct in_addr a4; 1248a915da9bSEric Dumazet #if IS_ENABLED(CONFIG_IPV6) 1249a915da9bSEric Dumazet struct in6_addr a6; 1250a915da9bSEric Dumazet #endif 1251a915da9bSEric Dumazet }; 1252a915da9bSEric Dumazet 1253cfb6eeb4SYOSHIFUJI Hideaki /* - key database */ 1254cfb6eeb4SYOSHIFUJI Hideaki struct tcp_md5sig_key { 1255a915da9bSEric Dumazet struct hlist_node node; 1256cfb6eeb4SYOSHIFUJI Hideaki u8 keylen; 1257a915da9bSEric Dumazet u8 family; /* AF_INET or AF_INET6 */ 1258a915da9bSEric Dumazet union tcp_md5_addr addr; 1259a915da9bSEric Dumazet u8 key[TCP_MD5SIG_MAXKEYLEN]; 1260a915da9bSEric Dumazet struct rcu_head rcu; 1261cfb6eeb4SYOSHIFUJI Hideaki }; 1262cfb6eeb4SYOSHIFUJI Hideaki 1263cfb6eeb4SYOSHIFUJI Hideaki /* - sock block */ 1264cfb6eeb4SYOSHIFUJI Hideaki struct tcp_md5sig_info { 1265a915da9bSEric Dumazet struct hlist_head head; 1266a8afca03SEric Dumazet struct rcu_head rcu; 1267cfb6eeb4SYOSHIFUJI Hideaki }; 1268cfb6eeb4SYOSHIFUJI Hideaki 1269cfb6eeb4SYOSHIFUJI Hideaki /* - pseudo header */ 1270cfb6eeb4SYOSHIFUJI Hideaki struct tcp4_pseudohdr { 1271cfb6eeb4SYOSHIFUJI Hideaki __be32 saddr; 1272cfb6eeb4SYOSHIFUJI Hideaki __be32 daddr; 1273cfb6eeb4SYOSHIFUJI Hideaki __u8 pad; 1274cfb6eeb4SYOSHIFUJI Hideaki __u8 protocol; 1275cfb6eeb4SYOSHIFUJI Hideaki __be16 len; 1276cfb6eeb4SYOSHIFUJI Hideaki }; 1277cfb6eeb4SYOSHIFUJI Hideaki 1278cfb6eeb4SYOSHIFUJI Hideaki struct tcp6_pseudohdr { 1279cfb6eeb4SYOSHIFUJI Hideaki struct in6_addr saddr; 1280cfb6eeb4SYOSHIFUJI Hideaki struct in6_addr daddr; 1281cfb6eeb4SYOSHIFUJI Hideaki __be32 len; 1282cfb6eeb4SYOSHIFUJI Hideaki __be32 protocol; /* including padding */ 1283cfb6eeb4SYOSHIFUJI Hideaki }; 1284cfb6eeb4SYOSHIFUJI Hideaki 1285cfb6eeb4SYOSHIFUJI Hideaki union tcp_md5sum_block { 1286cfb6eeb4SYOSHIFUJI Hideaki struct tcp4_pseudohdr ip4; 1287dfd56b8bSEric Dumazet #if IS_ENABLED(CONFIG_IPV6) 1288cfb6eeb4SYOSHIFUJI Hideaki struct tcp6_pseudohdr ip6; 1289cfb6eeb4SYOSHIFUJI Hideaki #endif 1290cfb6eeb4SYOSHIFUJI Hideaki }; 1291cfb6eeb4SYOSHIFUJI Hideaki 1292cfb6eeb4SYOSHIFUJI Hideaki /* - pool: digest algorithm, hash description and scratch buffer */ 1293cfb6eeb4SYOSHIFUJI Hideaki struct tcp_md5sig_pool { 1294cfb6eeb4SYOSHIFUJI Hideaki struct hash_desc md5_desc; 1295cfb6eeb4SYOSHIFUJI Hideaki union tcp_md5sum_block md5_blk; 1296cfb6eeb4SYOSHIFUJI Hideaki }; 1297cfb6eeb4SYOSHIFUJI Hideaki 1298cfb6eeb4SYOSHIFUJI Hideaki /* - functions */ 129939f8e58eSEric Dumazet int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, 130039f8e58eSEric Dumazet const struct sock *sk, const struct sk_buff *skb); 13015c9f3023SJoe Perches int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, 13025c9f3023SJoe Perches int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); 13035c9f3023SJoe Perches int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, 1304a915da9bSEric Dumazet int family); 13055c9f3023SJoe Perches struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, 1306fd3a154aSEric Dumazet const struct sock *addr_sk); 1307cfb6eeb4SYOSHIFUJI Hideaki 13089501f972SYOSHIFUJI Hideaki #ifdef CONFIG_TCP_MD5SIG 13095c9f3023SJoe Perches struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 13105c9f3023SJoe Perches const union tcp_md5_addr *addr, 13115c9f3023SJoe Perches int family); 1312a915da9bSEric Dumazet #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) 13139501f972SYOSHIFUJI Hideaki #else 1314a915da9bSEric Dumazet static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, 1315a915da9bSEric Dumazet const union tcp_md5_addr *addr, 1316a915da9bSEric Dumazet int family) 1317a915da9bSEric Dumazet { 1318a915da9bSEric Dumazet return NULL; 1319a915da9bSEric Dumazet } 13209501f972SYOSHIFUJI Hideaki #define tcp_twsk_md5_key(twsk) NULL 13219501f972SYOSHIFUJI Hideaki #endif 13229501f972SYOSHIFUJI Hideaki 13235c9f3023SJoe Perches bool tcp_alloc_md5sig_pool(void); 1324cfb6eeb4SYOSHIFUJI Hideaki 13255c9f3023SJoe Perches struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); 132671cea17eSEric Dumazet static inline void tcp_put_md5sig_pool(void) 132771cea17eSEric Dumazet { 132871cea17eSEric Dumazet local_bh_enable(); 132971cea17eSEric Dumazet } 133035790c04SEric Dumazet 13315c9f3023SJoe Perches int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); 13325c9f3023SJoe Perches int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, 133395c96174SEric Dumazet unsigned int header_len); 13345c9f3023SJoe Perches int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, 1335cf533ea5SEric Dumazet const struct tcp_md5sig_key *key); 1336cfb6eeb4SYOSHIFUJI Hideaki 133710467163SJerry Chu /* From tcp_fastopen.c */ 13385c9f3023SJoe Perches void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, 13395c9f3023SJoe Perches struct tcp_fastopen_cookie *cookie, int *syn_loss, 13405c9f3023SJoe Perches unsigned long *last_syn_loss); 13415c9f3023SJoe Perches void tcp_fastopen_cache_set(struct sock *sk, u16 mss, 1342*2646c831SDaniel Lee struct tcp_fastopen_cookie *cookie, bool syn_lost, 1343*2646c831SDaniel Lee u16 try_exp); 1344783237e8SYuchung Cheng struct tcp_fastopen_request { 1345783237e8SYuchung Cheng /* Fast Open cookie. Size 0 means a cookie request */ 1346783237e8SYuchung Cheng struct tcp_fastopen_cookie cookie; 1347783237e8SYuchung Cheng struct msghdr *data; /* data in MSG_FASTOPEN */ 1348f5ddcbbbSEric Dumazet size_t size; 1349f5ddcbbbSEric Dumazet int copied; /* queued in tcp_connect() */ 1350783237e8SYuchung Cheng }; 1351783237e8SYuchung Cheng void tcp_free_fastopen_req(struct tcp_sock *tp); 1352783237e8SYuchung Cheng 135310467163SJerry Chu extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; 135410467163SJerry Chu int tcp_fastopen_reset_cipher(void *key, unsigned int len); 1355843f4a55SYuchung Cheng bool tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, 13565b7ed089SYuchung Cheng struct request_sock *req, 1357843f4a55SYuchung Cheng struct tcp_fastopen_cookie *foc, 1358843f4a55SYuchung Cheng struct dst_entry *dst); 1359222e83d2SHannes Frederic Sowa void tcp_fastopen_init_key_once(bool publish); 136010467163SJerry Chu #define TCP_FASTOPEN_KEY_LENGTH 16 136110467163SJerry Chu 136210467163SJerry Chu /* Fastopen key context */ 136310467163SJerry Chu struct tcp_fastopen_context { 13647ae8639cSEric Dumazet struct crypto_cipher *tfm; 136510467163SJerry Chu __u8 key[TCP_FASTOPEN_KEY_LENGTH]; 136610467163SJerry Chu struct rcu_head rcu; 136710467163SJerry Chu }; 136810467163SJerry Chu 1369fe067e8aSDavid S. Miller /* write queue abstraction */ 1370fe067e8aSDavid S. Miller static inline void tcp_write_queue_purge(struct sock *sk) 1371fe067e8aSDavid S. Miller { 1372fe067e8aSDavid S. Miller struct sk_buff *skb; 1373fe067e8aSDavid S. Miller 1374fe067e8aSDavid S. Miller while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) 13753ab224beSHideo Aoki sk_wmem_free_skb(sk, skb); 13763ab224beSHideo Aoki sk_mem_reclaim(sk); 13778818a9d8SIlpo Järvinen tcp_clear_all_retrans_hints(tcp_sk(sk)); 1378fe067e8aSDavid S. Miller } 1379fe067e8aSDavid S. Miller 1380cf533ea5SEric Dumazet static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) 1381fe067e8aSDavid S. Miller { 1382cd07a8eaSDavid S. Miller return skb_peek(&sk->sk_write_queue); 1383fe067e8aSDavid S. Miller } 1384fe067e8aSDavid S. Miller 1385cf533ea5SEric Dumazet static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) 1386fe067e8aSDavid S. Miller { 1387cd07a8eaSDavid S. Miller return skb_peek_tail(&sk->sk_write_queue); 1388fe067e8aSDavid S. Miller } 1389fe067e8aSDavid S. Miller 1390cf533ea5SEric Dumazet static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, 1391cf533ea5SEric Dumazet const struct sk_buff *skb) 1392fe067e8aSDavid S. Miller { 1393cd07a8eaSDavid S. Miller return skb_queue_next(&sk->sk_write_queue, skb); 1394fe067e8aSDavid S. Miller } 1395fe067e8aSDavid S. Miller 1396cf533ea5SEric Dumazet static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, 1397cf533ea5SEric Dumazet const struct sk_buff *skb) 1398832d11c5SIlpo Järvinen { 1399832d11c5SIlpo Järvinen return skb_queue_prev(&sk->sk_write_queue, skb); 1400832d11c5SIlpo Järvinen } 1401832d11c5SIlpo Järvinen 1402fe067e8aSDavid S. Miller #define tcp_for_write_queue(skb, sk) \ 1403cd07a8eaSDavid S. Miller skb_queue_walk(&(sk)->sk_write_queue, skb) 1404fe067e8aSDavid S. Miller 1405fe067e8aSDavid S. Miller #define tcp_for_write_queue_from(skb, sk) \ 1406cd07a8eaSDavid S. Miller skb_queue_walk_from(&(sk)->sk_write_queue, skb) 1407fe067e8aSDavid S. Miller 1408234b6860SIlpo Järvinen #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ 1409cd07a8eaSDavid S. Miller skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) 1410234b6860SIlpo Järvinen 1411cf533ea5SEric Dumazet static inline struct sk_buff *tcp_send_head(const struct sock *sk) 1412fe067e8aSDavid S. Miller { 1413fe067e8aSDavid S. Miller return sk->sk_send_head; 1414fe067e8aSDavid S. Miller } 1415fe067e8aSDavid S. Miller 1416cd07a8eaSDavid S. Miller static inline bool tcp_skb_is_last(const struct sock *sk, 1417cd07a8eaSDavid S. Miller const struct sk_buff *skb) 1418cd07a8eaSDavid S. Miller { 1419cd07a8eaSDavid S. Miller return skb_queue_is_last(&sk->sk_write_queue, skb); 1420cd07a8eaSDavid S. Miller } 1421cd07a8eaSDavid S. Miller 1422cf533ea5SEric Dumazet static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) 1423fe067e8aSDavid S. Miller { 1424cd07a8eaSDavid S. Miller if (tcp_skb_is_last(sk, skb)) 1425fe067e8aSDavid S. Miller sk->sk_send_head = NULL; 1426cd07a8eaSDavid S. Miller else 1427cd07a8eaSDavid S. Miller sk->sk_send_head = tcp_write_queue_next(sk, skb); 1428fe067e8aSDavid S. Miller } 1429fe067e8aSDavid S. Miller 1430fe067e8aSDavid S. Miller static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) 1431fe067e8aSDavid S. Miller { 1432fe067e8aSDavid S. Miller if (sk->sk_send_head == skb_unlinked) 1433fe067e8aSDavid S. Miller sk->sk_send_head = NULL; 1434fe067e8aSDavid S. Miller } 1435fe067e8aSDavid S. Miller 1436fe067e8aSDavid S. Miller static inline void tcp_init_send_head(struct sock *sk) 1437fe067e8aSDavid S. Miller { 1438fe067e8aSDavid S. Miller sk->sk_send_head = NULL; 1439fe067e8aSDavid S. Miller } 1440fe067e8aSDavid S. Miller 1441fe067e8aSDavid S. Miller static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1442fe067e8aSDavid S. Miller { 1443fe067e8aSDavid S. Miller __skb_queue_tail(&sk->sk_write_queue, skb); 1444fe067e8aSDavid S. Miller } 1445fe067e8aSDavid S. Miller 1446fe067e8aSDavid S. Miller static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) 1447fe067e8aSDavid S. Miller { 1448fe067e8aSDavid S. Miller __tcp_add_write_queue_tail(sk, skb); 1449fe067e8aSDavid S. Miller 1450fe067e8aSDavid S. Miller /* Queue it, remembering where we must start sending. */ 14516859d494SIlpo Järvinen if (sk->sk_send_head == NULL) { 1452fe067e8aSDavid S. Miller sk->sk_send_head = skb; 14536859d494SIlpo Järvinen 14546859d494SIlpo Järvinen if (tcp_sk(sk)->highest_sack == NULL) 14556859d494SIlpo Järvinen tcp_sk(sk)->highest_sack = skb; 14566859d494SIlpo Järvinen } 1457fe067e8aSDavid S. Miller } 1458fe067e8aSDavid S. Miller 1459fe067e8aSDavid S. Miller static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) 1460fe067e8aSDavid S. Miller { 1461fe067e8aSDavid S. Miller __skb_queue_head(&sk->sk_write_queue, skb); 1462fe067e8aSDavid S. Miller } 1463fe067e8aSDavid S. Miller 1464fe067e8aSDavid S. Miller /* Insert buff after skb on the write queue of sk. */ 1465fe067e8aSDavid S. Miller static inline void tcp_insert_write_queue_after(struct sk_buff *skb, 1466fe067e8aSDavid S. Miller struct sk_buff *buff, 1467fe067e8aSDavid S. Miller struct sock *sk) 1468fe067e8aSDavid S. Miller { 14697de6c033SGerrit Renker __skb_queue_after(&sk->sk_write_queue, skb, buff); 1470fe067e8aSDavid S. Miller } 1471fe067e8aSDavid S. Miller 147243f59c89SDavid S. Miller /* Insert new before skb on the write queue of sk. */ 1473fe067e8aSDavid S. Miller static inline void tcp_insert_write_queue_before(struct sk_buff *new, 1474fe067e8aSDavid S. Miller struct sk_buff *skb, 1475fe067e8aSDavid S. Miller struct sock *sk) 1476fe067e8aSDavid S. Miller { 147743f59c89SDavid S. Miller __skb_queue_before(&sk->sk_write_queue, skb, new); 14786e421410SIlpo Järvinen 14796e421410SIlpo Järvinen if (sk->sk_send_head == skb) 14806e421410SIlpo Järvinen sk->sk_send_head = new; 1481fe067e8aSDavid S. Miller } 1482fe067e8aSDavid S. Miller 1483fe067e8aSDavid S. Miller static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) 1484fe067e8aSDavid S. Miller { 1485fe067e8aSDavid S. Miller __skb_unlink(skb, &sk->sk_write_queue); 1486fe067e8aSDavid S. Miller } 1487fe067e8aSDavid S. Miller 1488a2a385d6SEric Dumazet static inline bool tcp_write_queue_empty(struct sock *sk) 1489fe067e8aSDavid S. Miller { 1490fe067e8aSDavid S. Miller return skb_queue_empty(&sk->sk_write_queue); 1491fe067e8aSDavid S. Miller } 1492fe067e8aSDavid S. Miller 149312d50c46SKrishna Kumar static inline void tcp_push_pending_frames(struct sock *sk) 149412d50c46SKrishna Kumar { 149512d50c46SKrishna Kumar if (tcp_send_head(sk)) { 149612d50c46SKrishna Kumar struct tcp_sock *tp = tcp_sk(sk); 149712d50c46SKrishna Kumar 149812d50c46SKrishna Kumar __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); 149912d50c46SKrishna Kumar } 150012d50c46SKrishna Kumar } 150112d50c46SKrishna Kumar 1502ecb97192SNeal Cardwell /* Start sequence of the skb just after the highest skb with SACKed 1503ecb97192SNeal Cardwell * bit, valid only if sacked_out > 0 or when the caller has ensured 1504ecb97192SNeal Cardwell * validity by itself. 1505a47e5a98SIlpo Järvinen */ 1506a47e5a98SIlpo Järvinen static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) 1507a47e5a98SIlpo Järvinen { 1508a47e5a98SIlpo Järvinen if (!tp->sacked_out) 1509a47e5a98SIlpo Järvinen return tp->snd_una; 15106859d494SIlpo Järvinen 15116859d494SIlpo Järvinen if (tp->highest_sack == NULL) 15126859d494SIlpo Järvinen return tp->snd_nxt; 15136859d494SIlpo Järvinen 1514a47e5a98SIlpo Järvinen return TCP_SKB_CB(tp->highest_sack)->seq; 1515a47e5a98SIlpo Järvinen } 1516a47e5a98SIlpo Järvinen 15176859d494SIlpo Järvinen static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) 15186859d494SIlpo Järvinen { 15196859d494SIlpo Järvinen tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : 15206859d494SIlpo Järvinen tcp_write_queue_next(sk, skb); 15216859d494SIlpo Järvinen } 15226859d494SIlpo Järvinen 15236859d494SIlpo Järvinen static inline struct sk_buff *tcp_highest_sack(struct sock *sk) 15246859d494SIlpo Järvinen { 15256859d494SIlpo Järvinen return tcp_sk(sk)->highest_sack; 15266859d494SIlpo Järvinen } 15276859d494SIlpo Järvinen 15286859d494SIlpo Järvinen static inline void tcp_highest_sack_reset(struct sock *sk) 15296859d494SIlpo Järvinen { 15306859d494SIlpo Järvinen tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); 15316859d494SIlpo Järvinen } 15326859d494SIlpo Järvinen 15336859d494SIlpo Järvinen /* Called when old skb is about to be deleted (to be combined with new skb) */ 15346859d494SIlpo Järvinen static inline void tcp_highest_sack_combine(struct sock *sk, 15356859d494SIlpo Järvinen struct sk_buff *old, 15366859d494SIlpo Järvinen struct sk_buff *new) 15376859d494SIlpo Järvinen { 15386859d494SIlpo Järvinen if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) 15396859d494SIlpo Järvinen tcp_sk(sk)->highest_sack = new; 15406859d494SIlpo Järvinen } 15416859d494SIlpo Järvinen 15425aa4b32fSAndreas Petlund /* Determines whether this is a thin stream (which may suffer from 15435aa4b32fSAndreas Petlund * increased latency). Used to trigger latency-reducing mechanisms. 15445aa4b32fSAndreas Petlund */ 1545a2a385d6SEric Dumazet static inline bool tcp_stream_is_thin(struct tcp_sock *tp) 15465aa4b32fSAndreas Petlund { 15475aa4b32fSAndreas Petlund return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); 15485aa4b32fSAndreas Petlund } 15495aa4b32fSAndreas Petlund 15501da177e4SLinus Torvalds /* /proc */ 15511da177e4SLinus Torvalds enum tcp_seq_states { 15521da177e4SLinus Torvalds TCP_SEQ_STATE_LISTENING, 15531da177e4SLinus Torvalds TCP_SEQ_STATE_OPENREQ, 15541da177e4SLinus Torvalds TCP_SEQ_STATE_ESTABLISHED, 15551da177e4SLinus Torvalds }; 15561da177e4SLinus Torvalds 155773cb88ecSArjan van de Ven int tcp_seq_open(struct inode *inode, struct file *file); 155873cb88ecSArjan van de Ven 15591da177e4SLinus Torvalds struct tcp_seq_afinfo { 15601da177e4SLinus Torvalds char *name; 15611da177e4SLinus Torvalds sa_family_t family; 156273cb88ecSArjan van de Ven const struct file_operations *seq_fops; 15639427c4b3SDenis V. Lunev struct seq_operations seq_ops; 15641da177e4SLinus Torvalds }; 15651da177e4SLinus Torvalds 15661da177e4SLinus Torvalds struct tcp_iter_state { 1567a4146b1bSDenis V. Lunev struct seq_net_private p; 15681da177e4SLinus Torvalds sa_family_t family; 15691da177e4SLinus Torvalds enum tcp_seq_states state; 15701da177e4SLinus Torvalds struct sock *syn_wait_sk; 1571a7cb5a49SEric W. Biederman int bucket, offset, sbucket, num; 1572a7cb5a49SEric W. Biederman kuid_t uid; 1573a8b690f9STom Herbert loff_t last_pos; 15741da177e4SLinus Torvalds }; 15751da177e4SLinus Torvalds 15765c9f3023SJoe Perches int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); 15775c9f3023SJoe Perches void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); 15781da177e4SLinus Torvalds 157920380731SArnaldo Carvalho de Melo extern struct request_sock_ops tcp_request_sock_ops; 1580c6aefafbSGlenn Griffin extern struct request_sock_ops tcp6_request_sock_ops; 158120380731SArnaldo Carvalho de Melo 15825c9f3023SJoe Perches void tcp_v4_destroy_sock(struct sock *sk); 158320380731SArnaldo Carvalho de Melo 158428be6e07SEric Dumazet struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 1585c8f44affSMichał Mirosław netdev_features_t features); 15865c9f3023SJoe Perches struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); 15875c9f3023SJoe Perches int tcp_gro_complete(struct sk_buff *skb); 158828850dc7SDaniel Borkmann 15895c9f3023SJoe Perches void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); 1590f4c50d99SHerbert Xu 1591c9bee3b7SEric Dumazet static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) 1592c9bee3b7SEric Dumazet { 1593c9bee3b7SEric Dumazet return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat; 1594c9bee3b7SEric Dumazet } 1595c9bee3b7SEric Dumazet 1596c9bee3b7SEric Dumazet static inline bool tcp_stream_memory_free(const struct sock *sk) 1597c9bee3b7SEric Dumazet { 1598c9bee3b7SEric Dumazet const struct tcp_sock *tp = tcp_sk(sk); 1599c9bee3b7SEric Dumazet u32 notsent_bytes = tp->write_seq - tp->snd_nxt; 1600c9bee3b7SEric Dumazet 1601c9bee3b7SEric Dumazet return notsent_bytes < tcp_notsent_lowat(tp); 1602c9bee3b7SEric Dumazet } 1603c9bee3b7SEric Dumazet 160420380731SArnaldo Carvalho de Melo #ifdef CONFIG_PROC_FS 16055c9f3023SJoe Perches int tcp4_proc_init(void); 16065c9f3023SJoe Perches void tcp4_proc_exit(void); 160720380731SArnaldo Carvalho de Melo #endif 160820380731SArnaldo Carvalho de Melo 16095db92c99SOctavian Purdila int tcp_rtx_synack(struct sock *sk, struct request_sock *req); 16101fb6f159SOctavian Purdila int tcp_conn_request(struct request_sock_ops *rsk_ops, 16111fb6f159SOctavian Purdila const struct tcp_request_sock_ops *af_ops, 16121fb6f159SOctavian Purdila struct sock *sk, struct sk_buff *skb); 16135db92c99SOctavian Purdila 1614cfb6eeb4SYOSHIFUJI Hideaki /* TCP af-specific functions */ 1615cfb6eeb4SYOSHIFUJI Hideaki struct tcp_sock_af_ops { 1616cfb6eeb4SYOSHIFUJI Hideaki #ifdef CONFIG_TCP_MD5SIG 1617cfb6eeb4SYOSHIFUJI Hideaki struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk, 1618fd3a154aSEric Dumazet const struct sock *addr_sk); 1619cfb6eeb4SYOSHIFUJI Hideaki int (*calc_md5_hash)(char *location, 162039f8e58eSEric Dumazet const struct tcp_md5sig_key *md5, 1621318cf7aaSEric Dumazet const struct sock *sk, 1622318cf7aaSEric Dumazet const struct sk_buff *skb); 1623cfb6eeb4SYOSHIFUJI Hideaki int (*md5_parse)(struct sock *sk, 1624cfb6eeb4SYOSHIFUJI Hideaki char __user *optval, 1625cfb6eeb4SYOSHIFUJI Hideaki int optlen); 1626cfb6eeb4SYOSHIFUJI Hideaki #endif 1627cfb6eeb4SYOSHIFUJI Hideaki }; 1628cfb6eeb4SYOSHIFUJI Hideaki 1629cfb6eeb4SYOSHIFUJI Hideaki struct tcp_request_sock_ops { 16302aec4a29SOctavian Purdila u16 mss_clamp; 1631cfb6eeb4SYOSHIFUJI Hideaki #ifdef CONFIG_TCP_MD5SIG 1632fd3a154aSEric Dumazet struct tcp_md5sig_key *(*req_md5_lookup)(struct sock *sk, 1633fd3a154aSEric Dumazet const struct sock *addr_sk); 1634e3afe7b7SJohn Dykstra int (*calc_md5_hash) (char *location, 163539f8e58eSEric Dumazet const struct tcp_md5sig_key *md5, 1636318cf7aaSEric Dumazet const struct sock *sk, 1637318cf7aaSEric Dumazet const struct sk_buff *skb); 1638cfb6eeb4SYOSHIFUJI Hideaki #endif 163916bea70aSOctavian Purdila void (*init_req)(struct request_sock *req, struct sock *sk, 164016bea70aSOctavian Purdila struct sk_buff *skb); 1641fb7b37a7SOctavian Purdila #ifdef CONFIG_SYN_COOKIES 1642fb7b37a7SOctavian Purdila __u32 (*cookie_init_seq)(struct sock *sk, const struct sk_buff *skb, 1643fb7b37a7SOctavian Purdila __u16 *mss); 1644fb7b37a7SOctavian Purdila #endif 1645d94e0417SOctavian Purdila struct dst_entry *(*route_req)(struct sock *sk, struct flowi *fl, 1646d94e0417SOctavian Purdila const struct request_sock *req, 1647d94e0417SOctavian Purdila bool *strict); 1648936b8bdbSOctavian Purdila __u32 (*init_seq)(const struct sk_buff *skb); 1649d6274bd8SOctavian Purdila int (*send_synack)(struct sock *sk, struct dst_entry *dst, 1650d6274bd8SOctavian Purdila struct flowi *fl, struct request_sock *req, 1651d6274bd8SOctavian Purdila u16 queue_mapping, struct tcp_fastopen_cookie *foc); 1652695da14eSOctavian Purdila void (*queue_hash_add)(struct sock *sk, struct request_sock *req, 1653695da14eSOctavian Purdila const unsigned long timeout); 1654cfb6eeb4SYOSHIFUJI Hideaki }; 1655cfb6eeb4SYOSHIFUJI Hideaki 1656fb7b37a7SOctavian Purdila #ifdef CONFIG_SYN_COOKIES 1657fb7b37a7SOctavian Purdila static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1658fb7b37a7SOctavian Purdila struct sock *sk, struct sk_buff *skb, 1659fb7b37a7SOctavian Purdila __u16 *mss) 1660fb7b37a7SOctavian Purdila { 1661fb7b37a7SOctavian Purdila return ops->cookie_init_seq(sk, skb, mss); 1662fb7b37a7SOctavian Purdila } 1663fb7b37a7SOctavian Purdila #else 1664fb7b37a7SOctavian Purdila static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, 1665fb7b37a7SOctavian Purdila struct sock *sk, struct sk_buff *skb, 1666fb7b37a7SOctavian Purdila __u16 *mss) 1667fb7b37a7SOctavian Purdila { 1668fb7b37a7SOctavian Purdila return 0; 1669fb7b37a7SOctavian Purdila } 1670fb7b37a7SOctavian Purdila #endif 1671fb7b37a7SOctavian Purdila 16725c9f3023SJoe Perches int tcpv4_offload_init(void); 167328850dc7SDaniel Borkmann 16745c9f3023SJoe Perches void tcp_v4_init(void); 16755c9f3023SJoe Perches void tcp_init(void); 167620380731SArnaldo Carvalho de Melo 1677e25f866fSCong Wang /* 1678e25f866fSCong Wang * Save and compile IPv4 options, return a pointer to it 1679e25f866fSCong Wang */ 1680e25f866fSCong Wang static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) 1681e25f866fSCong Wang { 1682e25f866fSCong Wang const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 1683e25f866fSCong Wang struct ip_options_rcu *dopt = NULL; 1684e25f866fSCong Wang 1685461b74c3SCong Wang if (opt->optlen) { 1686e25f866fSCong Wang int opt_size = sizeof(*dopt) + opt->optlen; 1687e25f866fSCong Wang 1688e25f866fSCong Wang dopt = kmalloc(opt_size, GFP_ATOMIC); 1689e25f866fSCong Wang if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { 1690e25f866fSCong Wang kfree(dopt); 1691e25f866fSCong Wang dopt = NULL; 1692e25f866fSCong Wang } 1693e25f866fSCong Wang } 1694e25f866fSCong Wang return dopt; 1695e25f866fSCong Wang } 1696e25f866fSCong Wang 169798781965SEric Dumazet /* locally generated TCP pure ACKs have skb->truesize == 2 169898781965SEric Dumazet * (check tcp_send_ack() in net/ipv4/tcp_output.c ) 169998781965SEric Dumazet * This is much faster than dissecting the packet to find out. 170098781965SEric Dumazet * (Think of GRE encapsulations, IPv4, IPv6, ...) 170198781965SEric Dumazet */ 170298781965SEric Dumazet static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) 170398781965SEric Dumazet { 170498781965SEric Dumazet return skb->truesize == 2; 170598781965SEric Dumazet } 170698781965SEric Dumazet 170798781965SEric Dumazet static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) 170898781965SEric Dumazet { 170998781965SEric Dumazet skb->truesize = 2; 171098781965SEric Dumazet } 171198781965SEric Dumazet 17121da177e4SLinus Torvalds #endif /* _TCP_H */ 1713