tcp.h (d0ed4c60abfb9a4ab6cd416d1dea9df6266f8fc7) tcp.h (40a1227ea845a37ab197dd1caffb60b047fa36b1)
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93

--- 22 unchanged lines hidden (view full) ---

31#include <linux/kref.h>
32#include <linux/ktime.h>
33
34#include <net/inet_connection_sock.h>
35#include <net/inet_timewait_sock.h>
36#include <net/inet_hashtables.h>
37#include <net/checksum.h>
38#include <net/request_sock.h>
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93

--- 22 unchanged lines hidden (view full) ---

31#include <linux/kref.h>
32#include <linux/ktime.h>
33
34#include <net/inet_connection_sock.h>
35#include <net/inet_timewait_sock.h>
36#include <net/inet_hashtables.h>
37#include <net/checksum.h>
38#include <net/request_sock.h>
39#include <net/sock_reuseport.h>
39#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
42#include <net/tcp_states.h>
43#include <net/inet_ecn.h>
44#include <net/dst.h>
45
46#include <linux/seq_file.h>

--- 421 unchanged lines hidden (view full) ---

468#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
469
470/* syncookies: remember time of last synqueue overflow
471 * But do not dirty this field too often (once per second is enough)
472 * It is racy as we do not hold a lock, but race is very minor.
473 */
474static inline void tcp_synq_overflow(const struct sock *sk)
475{
40#include <net/sock.h>
41#include <net/snmp.h>
42#include <net/ip.h>
43#include <net/tcp_states.h>
44#include <net/inet_ecn.h>
45#include <net/dst.h>
46
47#include <linux/seq_file.h>

--- 421 unchanged lines hidden (view full) ---

469#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
470
471/* syncookies: remember time of last synqueue overflow
472 * But do not dirty this field too often (once per second is enough)
473 * It is racy as we do not hold a lock, but race is very minor.
474 */
475static inline void tcp_synq_overflow(const struct sock *sk)
476{
476 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
477 unsigned long now = jiffies;
477 unsigned int last_overflow;
478 unsigned int now = jiffies;
478
479
479 if (time_after(now, last_overflow + HZ))
480 if (sk->sk_reuseport) {
481 struct sock_reuseport *reuse;
482
483 reuse = rcu_dereference(sk->sk_reuseport_cb);
484 if (likely(reuse)) {
485 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
486 if (time_after32(now, last_overflow + HZ))
487 WRITE_ONCE(reuse->synq_overflow_ts, now);
488 return;
489 }
490 }
491
492 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
493 if (time_after32(now, last_overflow + HZ))
480 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
481}
482
483/* syncookies: no recent synqueue overflow on this listening socket? */
484static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
485{
494 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
495}
496
497/* syncookies: no recent synqueue overflow on this listening socket? */
498static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
499{
486 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
500 unsigned int last_overflow;
501 unsigned int now = jiffies;
487
502
488 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
503 if (sk->sk_reuseport) {
504 struct sock_reuseport *reuse;
505
506 reuse = rcu_dereference(sk->sk_reuseport_cb);
507 if (likely(reuse)) {
508 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
509 return time_after32(now, last_overflow +
510 TCP_SYNCOOKIE_VALID);
511 }
512 }
513
514 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
515 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
489}
490
491static inline u32 tcp_cookie_time(void)
492{
493 u64 val = get_jiffies_64();
494
495 do_div(val, TCP_SYNCOOKIE_PERIOD);
496 return val;

--- 461 unchanged lines hidden (view full) ---

958 * setting cwnd and pacing rate.
959 * A sample is invalid if "delivered" or "interval_us" is negative.
960 */
961struct rate_sample {
962 u64 prior_mstamp; /* starting timestamp for interval */
963 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
964 s32 delivered; /* number of packets delivered over interval */
965 long interval_us; /* time for tp->delivered to incr "delivered" */
516}
517
518static inline u32 tcp_cookie_time(void)
519{
520 u64 val = get_jiffies_64();
521
522 do_div(val, TCP_SYNCOOKIE_PERIOD);
523 return val;

--- 461 unchanged lines hidden (view full) ---

985 * setting cwnd and pacing rate.
986 * A sample is invalid if "delivered" or "interval_us" is negative.
987 */
988struct rate_sample {
989 u64 prior_mstamp; /* starting timestamp for interval */
990 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
991 s32 delivered; /* number of packets delivered over interval */
992 long interval_us; /* time for tp->delivered to incr "delivered" */
993 u32 snd_interval_us; /* snd interval for delivered packets */
994 u32 rcv_interval_us; /* rcv interval for delivered packets */
966 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
967 int losses; /* number of packets marked lost upon ACK */
968 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
969 u32 prior_in_flight; /* in flight before this ACK */
970 bool is_app_limited; /* is sample from packet with bubble in pipe? */
971 bool is_retrans; /* is sample from retransmission? */
972 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
973};

--- 215 unchanged lines hidden (view full) ---

1189
1190 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1191 if (tcp_in_slow_start(tp))
1192 return tp->snd_cwnd < 2 * tp->max_packets_out;
1193
1194 return tp->is_cwnd_limited;
1195}
1196
995 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
996 int losses; /* number of packets marked lost upon ACK */
997 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
998 u32 prior_in_flight; /* in flight before this ACK */
999 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1000 bool is_retrans; /* is sample from retransmission? */
1001 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1002};

--- 215 unchanged lines hidden (view full) ---

1218
1219 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1220 if (tcp_in_slow_start(tp))
1221 return tp->snd_cwnd < 2 * tp->max_packets_out;
1222
1223 return tp->is_cwnd_limited;
1224}
1225
1226/* BBR congestion control needs pacing.
1227 * Same remark for SO_MAX_PACING_RATE.
1228 * sch_fq packet scheduler is efficiently handling pacing,
1229 * but is not always installed/used.
1230 * Return true if TCP stack should pace packets itself.
1231 */
1232static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1233{
1234 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1235}
1236
1197/* Something is really bad, we could not queue an additional packet,
1198 * because qdisc is full or receiver sent a 0 window.
1199 * We do not want to add fuel to the fire, or abort too early,
1200 * so make sure the timer we arm now is at least 200ms in the future,
1201 * regardless of current icsk_rto value (as it could be ~2ms)
1202 */
1203static inline unsigned long tcp_probe0_base(const struct sock *sk)
1204{

--- 161 unchanged lines hidden (view full) ---

1366 return fin_timeout;
1367}
1368
1369static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1370 int paws_win)
1371{
1372 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1373 return true;
1237/* Something is really bad, we could not queue an additional packet,
1238 * because qdisc is full or receiver sent a 0 window.
1239 * We do not want to add fuel to the fire, or abort too early,
1240 * so make sure the timer we arm now is at least 200ms in the future,
1241 * regardless of current icsk_rto value (as it could be ~2ms)
1242 */
1243static inline unsigned long tcp_probe0_base(const struct sock *sk)
1244{

--- 161 unchanged lines hidden (view full) ---

1406 return fin_timeout;
1407}
1408
1409static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1410 int paws_win)
1411{
1412 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1413 return true;
1374 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1414 if (unlikely(!time_before32(ktime_get_seconds(),
1415 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1375 return true;
1376 /*
1377 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1378 * then following tcp messages have valid values. Ignore 0 value,
1379 * or else 'negative' tsval might forbid us to accept their packets.
1380 */
1381 if (!rx_opt->ts_recent)
1382 return true;

--- 13 unchanged lines hidden (view full) ---

1396 of this constraint to relax it: if peer reboots, clock may go
1397 out-of-sync and half-open connections will not be reset.
1398 Actually, the problem would be not existing if all
1399 the implementations followed draft about maintaining clock
1400 via reboots. Linux-2.2 DOES NOT!
1401
1402 However, we can relax time bounds for RST segments to MSL.
1403 */
1416 return true;
1417 /*
1418 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1419 * then following tcp messages have valid values. Ignore 0 value,
1420 * or else 'negative' tsval might forbid us to accept their packets.
1421 */
1422 if (!rx_opt->ts_recent)
1423 return true;

--- 13 unchanged lines hidden (view full) ---

1437 of this constraint to relax it: if peer reboots, clock may go
1438 out-of-sync and half-open connections will not be reset.
1439 Actually, the problem would be not existing if all
1440 the implementations followed draft about maintaining clock
1441 via reboots. Linux-2.2 DOES NOT!
1442
1443 However, we can relax time bounds for RST segments to MSL.
1444 */
1404 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1445 if (rst && !time_before32(ktime_get_seconds(),
1446 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1405 return false;
1406 return true;
1407}
1408
1409bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1410 int mib_idx, u32 *last_oow_ack_time);
1411
1412static inline void tcp_mib_init(struct net *net)

--- 369 unchanged lines hidden (view full) ---

1782
1783extern struct request_sock_ops tcp_request_sock_ops;
1784extern struct request_sock_ops tcp6_request_sock_ops;
1785
1786void tcp_v4_destroy_sock(struct sock *sk);
1787
1788struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1789 netdev_features_t features);
1447 return false;
1448 return true;
1449}
1450
1451bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1452 int mib_idx, u32 *last_oow_ack_time);
1453
1454static inline void tcp_mib_init(struct net *net)

--- 369 unchanged lines hidden (view full) ---

1824
1825extern struct request_sock_ops tcp_request_sock_ops;
1826extern struct request_sock_ops tcp6_request_sock_ops;
1827
1828void tcp_v4_destroy_sock(struct sock *sk);
1829
1830struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1831 netdev_features_t features);
1790struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1832struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1791int tcp_gro_complete(struct sk_buff *skb);
1792
1793void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1794
1795static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1796{
1797 struct net *net = sock_net((struct sock *)tp);
1798 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;

--- 330 unchanged lines hidden --- 		1833int tcp_gro_complete(struct sk_buff *skb);
1834
1835void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1836
1837static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1838{
1839 struct net *net = sock_net((struct sock *)tp);
1840 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;

--- 330 unchanged lines hidden ---