projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[NET]: {get|set}sockopt compatibility layer
[linux-block.git] / include / net / tcp.h
CommitLineData
1da177e4
LT		 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
9 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
21#define TCP_DEBUG 1
22#define FASTRETRANS_DEBUG 1
23
1da177e4
LT		 24#include <linux/config.h>
25#include <linux/list.h>
26#include <linux/tcp.h>
27#include <linux/slab.h>
28#include <linux/cache.h>
29#include <linux/percpu.h>
fb286bb2 30#include <linux/skbuff.h>
3f421baa
ACM		 31
32#include <net/inet_connection_sock.h>
295ff7ed 33#include <net/inet_timewait_sock.h>
77d8bf9c 34#include <net/inet_hashtables.h>
1da177e4 35#include <net/checksum.h>
2e6599cb 36#include <net/request_sock.h>
1da177e4
LT		 37#include <net/sock.h>
38#include <net/snmp.h>
39#include <net/ip.h>
c752f073
ACM		 40#include <net/tcp_states.h>
41
1da177e4
LT		 42#include <linux/seq_file.h>
43
6e04e021 44extern struct inet_hashinfo tcp_hashinfo;
1da177e4 45
1da177e4 46extern atomic_t tcp_orphan_count;
1da177e4 47extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 48
1da177e4
LT		 49#define MAX_TCP_HEADER (128 + MAX_HEADER)
50
51/*
52 * Never offer a window over 32767 without using window scaling. Some
53 * poor stacks do signed 16bit maths!
54 */
55#define MAX_TCP_WINDOW 32767U
56
57/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
58#define TCP_MIN_MSS 88U
59
60/* Minimal RCV_MSS. */
61#define TCP_MIN_RCVMSS 536U
62
5d424d5a
JH		 63/* The least MTU to use for probing */
64#define TCP_BASE_MSS 512
65
1da177e4
LT		 66/* After receiving this amount of duplicate ACKs fast retransmit starts. */
67#define TCP_FASTRETRANS_THRESH 3
68
69/* Maximal reordering. */
70#define TCP_MAX_REORDERING 127
71
72/* Maximal number of ACKs sent quickly to accelerate slow-start. */
73#define TCP_MAX_QUICKACKS 16U
74
75/* urg_data states */
76#define TCP_URG_VALID 0x0100
77#define TCP_URG_NOTYET 0x0200
78#define TCP_URG_READ 0x0400
79
80#define TCP_RETR1 3 /*
81 * This is how many retries it does before it
82 * tries to figure out if the gateway is
83 * down. Minimal RFC value is 3; it corresponds
84 * to ~3sec-8min depending on RTO.
85 */
86
87#define TCP_RETR2 15 /*
88 * This should take at least
89 * 90 minutes to time out.
90 * RFC1122 says that the limit is 100 sec.
91 * 15 is ~13-30min depending on RTO.
92 */
93
94#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
caa20d9a 95 * connection: ~180sec is RFC minimum */
1da177e4
LT		 96
97#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
caa20d9a 98 * connection: ~180sec is RFC minimum */
1da177e4
LT		 99
100
101#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
102 * socket. 7 is ~50sec-16min.
103 */
104
105
106#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
112 * TIME-WAIT timer.
113 */
114
115#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
116#if HZ >= 100
117#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118#define TCP_ATO_MIN ((unsigned)(HZ/25))
119#else
120#define TCP_DELACK_MIN 4U
121#define TCP_ATO_MIN 4U
122#endif
123#define TCP_RTO_MAX ((unsigned)(120*HZ))
124#define TCP_RTO_MIN ((unsigned)(HZ/5))
125#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
126
127#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
128 * for local resources.
129 */
130
131#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
132#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
133#define TCP_KEEPALIVE_INTVL (75*HZ)
134
135#define MAX_TCP_KEEPIDLE 32767
136#define MAX_TCP_KEEPINTVL 32767
137#define MAX_TCP_KEEPCNT 127
138#define MAX_TCP_SYNCNT 127
139
140#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
141#define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
142
143#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
144#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
145 * after this time. It should be equal
146 * (or greater than) TCP_TIMEWAIT_LEN
147 * to provide reliability equal to one
148 * provided by timewait state.
149 */
150#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
151 * timestamps. It must be less than
152 * minimal timewait lifetime.
153 */
1da177e4
LT		 154/*
155 * TCP option
156 */
157
158#define TCPOPT_NOP 1 /* Padding */
159#define TCPOPT_EOL 0 /* End of options */
160#define TCPOPT_MSS 2 /* Segment size negotiating */
161#define TCPOPT_WINDOW 3 /* Window scaling */
162#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
163#define TCPOPT_SACK 5 /* SACK Block */
164#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
165
166/*
167 * TCP option lengths
168 */
169
170#define TCPOLEN_MSS 4
171#define TCPOLEN_WINDOW 3
172#define TCPOLEN_SACK_PERM 2
173#define TCPOLEN_TIMESTAMP 10
174
175/* But this is what stacks really send out. */
176#define TCPOLEN_TSTAMP_ALIGNED 12
177#define TCPOLEN_WSCALE_ALIGNED 4
178#define TCPOLEN_SACKPERM_ALIGNED 4
179#define TCPOLEN_SACK_BASE 2
180#define TCPOLEN_SACK_BASE_ALIGNED 4
181#define TCPOLEN_SACK_PERBLOCK 8
182
1da177e4
LT		 183/* Flags in tp->nonagle */
184#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
185#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 186#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 187
295ff7ed
ACM		 188extern struct inet_timewait_death_row tcp_death_row;
189
1da177e4 190/* sysctl variables for tcp */
1da177e4
LT		 191extern int sysctl_tcp_timestamps;
192extern int sysctl_tcp_window_scaling;
193extern int sysctl_tcp_sack;
194extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 195extern int sysctl_tcp_keepalive_time;
196extern int sysctl_tcp_keepalive_probes;
197extern int sysctl_tcp_keepalive_intvl;
198extern int sysctl_tcp_syn_retries;
199extern int sysctl_tcp_synack_retries;
200extern int sysctl_tcp_retries1;
201extern int sysctl_tcp_retries2;
202extern int sysctl_tcp_orphan_retries;
203extern int sysctl_tcp_syncookies;
204extern int sysctl_tcp_retrans_collapse;
205extern int sysctl_tcp_stdurg;
206extern int sysctl_tcp_rfc1337;
207extern int sysctl_tcp_abort_on_overflow;
208extern int sysctl_tcp_max_orphans;
1da177e4
LT		 209extern int sysctl_tcp_fack;
210extern int sysctl_tcp_reordering;
211extern int sysctl_tcp_ecn;
212extern int sysctl_tcp_dsack;
213extern int sysctl_tcp_mem[3];
214extern int sysctl_tcp_wmem[3];
215extern int sysctl_tcp_rmem[3];
216extern int sysctl_tcp_app_win;
217extern int sysctl_tcp_adv_win_scale;
218extern int sysctl_tcp_tw_reuse;
219extern int sysctl_tcp_frto;
220extern int sysctl_tcp_low_latency;
1da177e4 221extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 222extern int sysctl_tcp_moderate_rcvbuf;
223extern int sysctl_tcp_tso_win_divisor;
9772efb9 224extern int sysctl_tcp_abc;
5d424d5a
JH		 225extern int sysctl_tcp_mtu_probing;
226extern int sysctl_tcp_base_mss;
15d99e02 227extern int sysctl_tcp_workaround_signed_windows;
1da177e4
LT		 228
229extern atomic_t tcp_memory_allocated;
230extern atomic_t tcp_sockets_allocated;
231extern int tcp_memory_pressure;
232
1da177e4
LT		 233/*
234 * The next routines deal with comparing 32 bit unsigned ints
235 * and worry about wraparound (automatic with unsigned arithmetic).
236 */
237
238static inline int before(__u32 seq1, __u32 seq2)
239{
240 return (__s32)(seq1-seq2) < 0;
241}
242
243static inline int after(__u32 seq1, __u32 seq2)
244{
245 return (__s32)(seq2-seq1) < 0;
246}
247
248
249/* is s2<=s1<=s3 ? */
250static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
251{
252 return seq3 - seq2 >= seq1 - seq2;
253}
254
255
256extern struct proto tcp_prot;
257
258DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
259#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
260#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
261#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
262#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
263#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
264#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
265
1da177e4
LT		 266extern void tcp_v4_err(struct sk_buff *skb, u32);
267
268extern void tcp_shutdown (struct sock *sk, int how);
269
270extern int tcp_v4_rcv(struct sk_buff *skb);
271
272extern int tcp_v4_remember_stamp(struct sock *sk);
273
8feaf0c0 274extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4
LT		 275
276extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
277 struct msghdr *msg, size_t size);
278extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
279
280extern int tcp_ioctl(struct sock *sk,
281 int cmd,
282 unsigned long arg);
283
284extern int tcp_rcv_state_process(struct sock *sk,
285 struct sk_buff *skb,
286 struct tcphdr *th,
287 unsigned len);
288
289extern int tcp_rcv_established(struct sock *sk,
290 struct sk_buff *skb,
291 struct tcphdr *th,
292 unsigned len);
293
294extern void tcp_rcv_space_adjust(struct sock *sk);
295
6d6ee43e
ACM		 296extern int tcp_twsk_unique(struct sock *sk,
297 struct sock *sktw, void *twp);
298
463c84b9
ACM		 299static inline void tcp_dec_quickack_mode(struct sock *sk,
300 const unsigned int pkts)
1da177e4 301{
463c84b9 302 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 303
463c84b9
ACM		 304 if (icsk->icsk_ack.quick) {
305 if (pkts >= icsk->icsk_ack.quick) {
306 icsk->icsk_ack.quick = 0;
fc6415bc 307 /* Leaving quickack mode we deflate ATO. */
463c84b9 308 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 309 } else
463c84b9 310 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 311 }
312}
313
463c84b9 314extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 315
1da177e4
LT		 316static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
317{
318 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
319}
320
321enum tcp_tw_status
322{
323 TCP_TW_SUCCESS = 0,
324 TCP_TW_RST = 1,
325 TCP_TW_ACK = 2,
326 TCP_TW_SYN = 3
327};
328
329
8feaf0c0 330extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 331 struct sk_buff *skb,
8feaf0c0 332 const struct tcphdr *th);
1da177e4
LT		 333
334extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 335 struct request_sock *req,
336 struct request_sock **prev);
1da177e4
LT		 337extern int tcp_child_process(struct sock *parent,
338 struct sock *child,
339 struct sk_buff *skb);
340extern void tcp_enter_frto(struct sock *sk);
341extern void tcp_enter_loss(struct sock *sk, int how);
342extern void tcp_clear_retrans(struct tcp_sock *tp);
343extern void tcp_update_metrics(struct sock *sk);
344
345extern void tcp_close(struct sock *sk,
346 long timeout);
1da177e4
LT		 347extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
348
349extern int tcp_getsockopt(struct sock *sk, int level,
350 int optname,
351 char __user *optval,
352 int __user *optlen);
353extern int tcp_setsockopt(struct sock *sk, int level,
354 int optname, char __user *optval,
355 int optlen);
3fdadf7d
DM		 356extern int compat_tcp_getsockopt(struct sock *sk,
357 int level, int optname,
358 char __user *optval, int __user *optlen);
359extern int compat_tcp_setsockopt(struct sock *sk,
360 int level, int optname,
361 char __user *optval, int optlen);
1da177e4
LT		 362extern void tcp_set_keepalive(struct sock *sk, int val);
363extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
364 struct msghdr *msg,
365 size_t len, int nonblock,
366 int flags, int *addr_len);
367
1da177e4
LT		 368extern void tcp_parse_options(struct sk_buff *skb,
369 struct tcp_options_received *opt_rx,
370 int estab);
371
372/*
373 * TCP v4 functions exported for the inet6 API
374 */
375
8292a17a 376extern void tcp_v4_send_check(struct sock *sk, int len,
1da177e4
LT		 377 struct sk_buff *skb);
378
379extern int tcp_v4_conn_request(struct sock *sk,
380 struct sk_buff *skb);
381
382extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 383 struct request_sock *req,
1da177e4
LT		 384 struct sk_buff *skb);
385
386extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
387 struct sk_buff *skb,
60236fdd 388 struct request_sock *req,
1da177e4
LT		 389 struct dst_entry *dst);
390
391extern int tcp_v4_do_rcv(struct sock *sk,
392 struct sk_buff *skb);
393
394extern int tcp_v4_connect(struct sock *sk,
395 struct sockaddr *uaddr,
396 int addr_len);
397
398extern int tcp_connect(struct sock *sk);
399
400extern struct sk_buff * tcp_make_synack(struct sock *sk,
401 struct dst_entry *dst,
60236fdd 402 struct request_sock *req);
1da177e4
LT		 403
404extern int tcp_disconnect(struct sock *sk, int flags);
405
406extern void tcp_unhash(struct sock *sk);
407
408extern int tcp_v4_hash_connecting(struct sock *sk);
409
410
411/* From syncookies.c */
412extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
413 struct ip_options *opt);
414extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
415 __u16 *mss);
416
417/* tcp_output.c */
418
f6302d1d 419extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
a2e2a59c 420 unsigned int cur_mss, int nonagle);
f6302d1d 421extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
1da177e4
LT		 422extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
423extern void tcp_xmit_retransmit_queue(struct sock *);
424extern void tcp_simple_retransmit(struct sock *);
425extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 426extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 427
428extern void tcp_send_probe0(struct sock *);
429extern void tcp_send_partial(struct sock *);
430extern int tcp_write_wakeup(struct sock *);
431extern void tcp_send_fin(struct sock *sk);
dd0fc66f 432extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
1da177e4 433extern int tcp_send_synack(struct sock *);
c1b4a7e6 434extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 435extern void tcp_send_ack(struct sock *sk);
436extern void tcp_send_delayed_ack(struct sock *sk);
437
a762a980
DM		 438/* tcp_input.c */
439extern void tcp_cwnd_application_limited(struct sock *sk);
440
1da177e4
LT		 441/* tcp_timer.c */
442extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 443static inline void tcp_clear_xmit_timers(struct sock *sk)
444{
445 inet_csk_clear_xmit_timers(sk);
446}
1da177e4 447
1da177e4
LT		 448extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
449extern unsigned int tcp_current_mss(struct sock *sk, int large);
450
17b085ea 451/* tcp.c */
1da177e4
LT		 452extern void tcp_get_info(struct sock *, struct tcp_info *);
453
454/* Read 'sendfile()'-style from a TCP socket */
455typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
456 unsigned int, size_t);
457extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
458 sk_read_actor_t recv_actor);
459
40efc6fa 460extern void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 461
5d424d5a
JH		 462extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
463extern int tcp_mss_to_mtu(struct sock *sk, int mss);
464extern void tcp_mtup_init(struct sock *sk);
465
40efc6fa 466static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 467{
468 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
469 ntohl(TCP_FLAG_ACK) |
470 snd_wnd);
471}
472
40efc6fa 473static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 474{
475 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
476}
477
478static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
479{
b03efcfb 480 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 481 tp->rcv_wnd &&
482 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
483 !tp->urg_data)
484 tcp_fast_path_on(tp);
485}
486
487/* Compute the actual receive window we are currently advertising.
488 * Rcv_nxt can be after the window if our peer push more data
489 * than the offered window.
490 */
40efc6fa 491static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 492{
493 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
494
495 if (win < 0)
496 win = 0;
497 return (u32) win;
498}
499
500/* Choose a new window, without checks for shrinking, and without
501 * scaling applied to the result. The caller does these things
502 * if necessary. This is a "raw" window selection.
503 */
504extern u32 __tcp_select_window(struct sock *sk);
505
506/* TCP timestamps are only 32-bits, this causes a slight
507 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 508 * of jiffies in the buffer control blocks below. We decided
509 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 510 * casts with the following macro.
511 */
512#define tcp_time_stamp ((__u32)(jiffies))
513
caa20d9a 514/* This is what the send packet queuing engine uses to pass
1da177e4
LT		 515 * TCP per-packet control information to the transmission
516 * code. We also store the host-order sequence numbers in
517 * here too. This is 36 bytes on 32-bit architectures,
518 * 40 bytes on 64-bit machines, if this grows please adjust
519 * skbuff.h:skbuff->cb[xxx] size appropriately.
520 */
521struct tcp_skb_cb {
522 union {
523 struct inet_skb_parm h4;
524#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
525 struct inet6_skb_parm h6;
526#endif
527 } header; /* For incoming frames */
528 __u32 seq; /* Starting sequence number */
529 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
530 __u32 when; /* used to compute rtt's */
531 __u8 flags; /* TCP header flags. */
532
533 /* NOTE: These must match up to the flags byte in a
534 * real TCP header.
535 */
536#define TCPCB_FLAG_FIN 0x01
537#define TCPCB_FLAG_SYN 0x02
538#define TCPCB_FLAG_RST 0x04
539#define TCPCB_FLAG_PSH 0x08
540#define TCPCB_FLAG_ACK 0x10
541#define TCPCB_FLAG_URG 0x20
542#define TCPCB_FLAG_ECE 0x40
543#define TCPCB_FLAG_CWR 0x80
544
545 __u8 sacked; /* State flags for SACK/FACK. */
546#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
547#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
548#define TCPCB_LOST 0x04 /* SKB is lost */
549#define TCPCB_TAGBITS 0x07 /* All tag bits */
550
551#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
552#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
553
caa20d9a 554#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
1da177e4
LT		 555
556#define TCPCB_AT_TAIL (TCPCB_URG)
557
558 __u16 urg_ptr; /* Valid w/URG flags is set. */
559 __u32 ack_seq; /* Sequence number ACK'd */
560};
561
562#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
563
564#include <net/tcp_ecn.h>
565
566/* Due to TSO, an SKB can be composed of multiple actual
567 * packets. To keep these tracked properly, we use this.
568 */
569static inline int tcp_skb_pcount(const struct sk_buff *skb)
570{
571 return skb_shinfo(skb)->tso_segs;
572}
573
574/* This is valid iff tcp_skb_pcount() > 1. */
575static inline int tcp_skb_mss(const struct sk_buff *skb)
576{
577 return skb_shinfo(skb)->tso_size;
578}
579
580static inline void tcp_dec_pcount_approx(__u32 *count,
581 const struct sk_buff *skb)
582{
583 if (*count) {
584 *count -= tcp_skb_pcount(skb);
585 if ((int)*count < 0)
586 *count = 0;
587 }
588}
589
590static inline void tcp_packets_out_inc(struct sock *sk,
591 struct tcp_sock *tp,
592 const struct sk_buff *skb)
593{
594 int orig = tp->packets_out;
595
596 tp->packets_out += tcp_skb_pcount(skb);
597 if (!orig)
3f421baa
ACM		 598 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
599 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 600}
601
602static inline void tcp_packets_out_dec(struct tcp_sock *tp,
603 const struct sk_buff *skb)
604{
605 tp->packets_out -= tcp_skb_pcount(skb);
606}
607
317a76f9
SH		 608/* Events passed to congestion control interface */
609enum tcp_ca_event {
610 CA_EVENT_TX_START, /* first transmit when no packets in flight */
611 CA_EVENT_CWND_RESTART, /* congestion window restart */
612 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
613 CA_EVENT_FRTO, /* fast recovery timeout */
614 CA_EVENT_LOSS, /* loss timeout */
615 CA_EVENT_FAST_ACK, /* in sequence ack */
616 CA_EVENT_SLOW_ACK, /* other ack */
617};
618
619/*
620 * Interface for adding new TCP congestion control handlers
621 */
622#define TCP_CA_NAME_MAX 16
623struct tcp_congestion_ops {
624 struct list_head list;
625
626 /* initialize private data (optional) */
6687e988 627 void (*init)(struct sock *sk);
317a76f9 628 /* cleanup private data (optional) */
6687e988 629 void (*release)(struct sock *sk);
317a76f9
SH		 630
631 /* return slow start threshold (required) */
6687e988 632 u32 (*ssthresh)(struct sock *sk);
317a76f9 633 /* lower bound for congestion window (optional) */
6687e988 634 u32 (*min_cwnd)(struct sock *sk);
317a76f9 635 /* do new cwnd calculation (required) */
6687e988 636 void (*cong_avoid)(struct sock *sk, u32 ack,
317a76f9
SH		 637 u32 rtt, u32 in_flight, int good_ack);
638 /* round trip time sample per acked packet (optional) */
6687e988 639 void (*rtt_sample)(struct sock *sk, u32 usrtt);
317a76f9 640 /* call before changing ca_state (optional) */
6687e988 641 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 642 /* call when cwnd event occurs (optional) */
6687e988 643 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 644 /* new value of cwnd after loss (optional) */
6687e988 645 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 646 /* hook for packet ack accounting (optional) */
6687e988 647 void (*pkts_acked)(struct sock *sk, u32 num_acked);
73c1f4a0 648 /* get info for inet_diag (optional) */
6687e988 649 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 650
651 char name[TCP_CA_NAME_MAX];
652 struct module *owner;
653};
654
655extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
656extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
657
6687e988
ACM		 658extern void tcp_init_congestion_control(struct sock *sk);
659extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 660extern int tcp_set_default_congestion_control(const char *name);
661extern void tcp_get_default_congestion_control(char *name);
6687e988 662extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 663extern void tcp_slow_start(struct tcp_sock *tp);
317a76f9 664
5f8ef48d 665extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988
ACM		 666extern u32 tcp_reno_ssthresh(struct sock *sk);
667extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
317a76f9 668 u32 rtt, u32 in_flight, int flag);
6687e988 669extern u32 tcp_reno_min_cwnd(struct sock *sk);
a8acfbac 670extern struct tcp_congestion_ops tcp_reno;
317a76f9 671
6687e988 672static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 673{
6687e988
ACM		 674 struct inet_connection_sock *icsk = inet_csk(sk);
675
676 if (icsk->icsk_ca_ops->set_state)
677 icsk->icsk_ca_ops->set_state(sk, ca_state);
678 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 679}
680
6687e988 681static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 682{
6687e988
ACM		 683 const struct inet_connection_sock *icsk = inet_csk(sk);
684
685 if (icsk->icsk_ca_ops->cwnd_event)
686 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 687}
688
1da177e4
LT		 689/* This determines how many packets are "in the network" to the best
690 * of our knowledge. In many cases it is conservative, but where
691 * detailed information is available from the receiver (via SACK
692 * blocks etc.) we can make more aggressive calculations.
693 *
694 * Use this for decisions involving congestion control, use just
695 * tp->packets_out to determine if the send queue is empty or not.
696 *
697 * Read this equation as:
698 *
699 * "Packets sent once on transmission queue" MINUS
700 * "Packets left network, but not honestly ACKed yet" PLUS
701 * "Packets fast retransmitted"
702 */
40efc6fa 703static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4
LT		 704{
705 return (tp->packets_out - tp->left_out + tp->retrans_out);
706}
707
1da177e4
LT		 708/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
709 * The exception is rate halving phase, when cwnd is decreasing towards
710 * ssthresh.
711 */
6687e988 712static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 713{
6687e988
ACM		 714 const struct tcp_sock *tp = tcp_sk(sk);
715 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 716 return tp->snd_ssthresh;
717 else
718 return max(tp->snd_ssthresh,
719 ((tp->snd_cwnd >> 1) +
720 (tp->snd_cwnd >> 2)));
721}
722
723static inline void tcp_sync_left_out(struct tcp_sock *tp)
724{
725 if (tp->rx_opt.sack_ok &&
726 (tp->sacked_out >= tp->packets_out - tp->lost_out))
727 tp->sacked_out = tp->packets_out - tp->lost_out;
728 tp->left_out = tp->sacked_out + tp->lost_out;
729}
730
40efc6fa 731extern void tcp_enter_cwr(struct sock *sk);
1da177e4
LT		 732extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
733
734/* Slow start with delack produces 3 packets of burst, so that
735 * it is safe "de facto".
736 */
737static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
738{
739 return 3;
740}
741
f4805ede
SH		 742/* RFC2861 Check whether we are limited by application or congestion window
743 * This is the inverse of cwnd check in tcp_tso_should_defer
744 */
745static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
746{
747 const struct tcp_sock *tp = tcp_sk(sk);
748 u32 left;
749
750 if (in_flight >= tp->snd_cwnd)
751 return 1;
752
753 if (!(sk->sk_route_caps & NETIF_F_TSO))
754 return 0;
755
756 left = tp->snd_cwnd - in_flight;
757 if (sysctl_tcp_tso_win_divisor)
758 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
759 else
760 return left <= tcp_max_burst(tp);
761}
762
40efc6fa
SH		 763static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
764 const struct sk_buff *skb)
1da177e4
LT		 765{
766 if (skb->len < mss)
767 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
768}
769
40efc6fa 770static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
1da177e4 771{
463c84b9
ACM		 772 const struct inet_connection_sock *icsk = inet_csk(sk);
773 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 774 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
775 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 776}
777
40efc6fa
SH		 778static inline void tcp_push_pending_frames(struct sock *sk,
779 struct tcp_sock *tp)
1da177e4
LT		 780{
781 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
782}
783
40efc6fa 784static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 785{
786 tp->snd_wl1 = seq;
787}
788
40efc6fa 789static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 790{
791 tp->snd_wl1 = seq;
792}
793
1da177e4
LT		 794/*
795 * Calculate(/check) TCP checksum
796 */
40efc6fa
SH		 797static inline u16 tcp_v4_check(struct tcphdr *th, int len,
798 unsigned long saddr, unsigned long daddr,
799 unsigned long base)
1da177e4
LT		 800{
801 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
802}
803
40efc6fa 804static inline int __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 805{
fb286bb2 806 return __skb_checksum_complete(skb);
1da177e4
LT		 807}
808
40efc6fa 809static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4
LT		 810{
811 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
812 __tcp_checksum_complete(skb);
813}
814
815/* Prequeue for VJ style copy to user, combined with checksumming. */
816
40efc6fa 817static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 818{
819 tp->ucopy.task = NULL;
820 tp->ucopy.len = 0;
821 tp->ucopy.memory = 0;
822 skb_queue_head_init(&tp->ucopy.prequeue);
823}
824
825/* Packet is added to VJ-style prequeue for processing in process
826 * context, if a reader task is waiting. Apparently, this exciting
827 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
828 * failed somewhere. Latency? Burstiness? Well, at least now we will
829 * see, why it failed. 8)8) --ANK
830 *
831 * NOTE: is this not too big to inline?
832 */
40efc6fa 833static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 834{
835 struct tcp_sock *tp = tcp_sk(sk);
836
837 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
838 __skb_queue_tail(&tp->ucopy.prequeue, skb);
839 tp->ucopy.memory += skb->truesize;
840 if (tp->ucopy.memory > sk->sk_rcvbuf) {
841 struct sk_buff *skb1;
842
843 BUG_ON(sock_owned_by_user(sk));
844
845 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
846 sk->sk_backlog_rcv(sk, skb1);
847 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
848 }
849
850 tp->ucopy.memory = 0;
851 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
852 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 853 if (!inet_csk_ack_scheduled(sk))
854 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
3f421baa
ACM		 855 (3 * TCP_RTO_MIN) / 4,
856 TCP_RTO_MAX);
1da177e4
LT		 857 }
858 return 1;
859 }
860 return 0;
861}
862
863
864#undef STATE_TRACE
865
866#ifdef STATE_TRACE
867static const char *statename[]={
868 "Unused","Established","Syn Sent","Syn Recv",
869 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
870 "Close Wait","Last ACK","Listen","Closing"
871};
872#endif
873
40efc6fa 874static inline void tcp_set_state(struct sock *sk, int state)
1da177e4
LT		 875{
876 int oldstate = sk->sk_state;
877
878 switch (state) {
879 case TCP_ESTABLISHED:
880 if (oldstate != TCP_ESTABLISHED)
881 TCP_INC_STATS(TCP_MIB_CURRESTAB);
882 break;
883
884 case TCP_CLOSE:
885 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
886 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
887
888 sk->sk_prot->unhash(sk);
463c84b9 889 if (inet_csk(sk)->icsk_bind_hash &&
1da177e4 890 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2d8c4ce5 891 inet_put_port(&tcp_hashinfo, sk);
1da177e4
LT		 892 /* fall through */
893 default:
894 if (oldstate==TCP_ESTABLISHED)
895 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
896 }
897
898 /* Change state AFTER socket is unhashed to avoid closed
899 * socket sitting in hash tables.
900 */
901 sk->sk_state = state;
902
903#ifdef STATE_TRACE
904 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
905#endif
906}
907
40efc6fa 908static inline void tcp_done(struct sock *sk)
1da177e4
LT		 909{
910 tcp_set_state(sk, TCP_CLOSE);
911 tcp_clear_xmit_timers(sk);
912
913 sk->sk_shutdown = SHUTDOWN_MASK;
914
915 if (!sock_flag(sk, SOCK_DEAD))
916 sk->sk_state_change(sk);
917 else
0a5578cf 918 inet_csk_destroy_sock(sk);
1da177e4
LT		 919}
920
40efc6fa 921static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 922{
923 rx_opt->dsack = 0;
924 rx_opt->eff_sacks = 0;
925 rx_opt->num_sacks = 0;
926}
927
1da177e4
LT		 928/* Determine a window scaling and initial window to offer. */
929extern void tcp_select_initial_window(int __space, __u32 mss,
930 __u32 *rcv_wnd, __u32 *window_clamp,
931 int wscale_ok, __u8 *rcv_wscale);
932
933static inline int tcp_win_from_space(int space)
934{
935 return sysctl_tcp_adv_win_scale<=0 ?
936 (space>>(-sysctl_tcp_adv_win_scale)) :
937 space - (space>>sysctl_tcp_adv_win_scale);
938}
939
940/* Note: caller must be prepared to deal with negative returns */
941static inline int tcp_space(const struct sock *sk)
942{
943 return tcp_win_from_space(sk->sk_rcvbuf -
944 atomic_read(&sk->sk_rmem_alloc));
945}
946
947static inline int tcp_full_space(const struct sock *sk)
948{
949 return tcp_win_from_space(sk->sk_rcvbuf);
950}
951
40efc6fa
SH		 952static inline void tcp_openreq_init(struct request_sock *req,
953 struct tcp_options_received *rx_opt,
954 struct sk_buff *skb)
1da177e4 955{
2e6599cb
ACM		 956 struct inet_request_sock *ireq = inet_rsk(req);
957
1da177e4 958 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 959 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 960 req->mss = rx_opt->mss_clamp;
961 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 962 ireq->tstamp_ok = rx_opt->tstamp_ok;
963 ireq->sack_ok = rx_opt->sack_ok;
964 ireq->snd_wscale = rx_opt->snd_wscale;
965 ireq->wscale_ok = rx_opt->wscale_ok;
966 ireq->acked = 0;
967 ireq->ecn_ok = 0;
968 ireq->rmt_port = skb->h.th->source;
1da177e4
LT		 969}
970
971extern void tcp_enter_memory_pressure(void);
972
1da177e4
LT		 973static inline int keepalive_intvl_when(const struct tcp_sock *tp)
974{
975 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
976}
977
978static inline int keepalive_time_when(const struct tcp_sock *tp)
979{
980 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
981}
982
463c84b9 983static inline int tcp_fin_time(const struct sock *sk)
1da177e4 984{
463c84b9
ACM		 985 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
986 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 987
463c84b9
ACM		 988 if (fin_timeout < (rto << 2) - (rto >> 1))
989 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 990
991 return fin_timeout;
992}
993
994static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
995{
996 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
997 return 0;
998 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
999 return 0;
1000
1001 /* RST segments are not recommended to carry timestamp,
1002 and, if they do, it is recommended to ignore PAWS because
1003 "their cleanup function should take precedence over timestamps."
1004 Certainly, it is mistake. It is necessary to understand the reasons
1005 of this constraint to relax it: if peer reboots, clock may go
1006 out-of-sync and half-open connections will not be reset.
1007 Actually, the problem would be not existing if all
1008 the implementations followed draft about maintaining clock
1009 via reboots. Linux-2.2 DOES NOT!
1010
1011 However, we can relax time bounds for RST segments to MSL.
1012 */
1013 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1014 return 0;
1015 return 1;
1016}
1017
1da177e4
LT		 1018#define TCP_CHECK_TIMER(sk) do { } while (0)
1019
1020static inline int tcp_use_frto(const struct sock *sk)
1021{
1022 const struct tcp_sock *tp = tcp_sk(sk);
1023
1024 /* F-RTO must be activated in sysctl and there must be some
1025 * unsent new data, and the advertised window should allow
1026 * sending it.
1027 */
1028 return (sysctl_tcp_frto && sk->sk_send_head &&
1029 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1030 tp->snd_una + tp->snd_wnd));
1031}
1032
1033static inline void tcp_mib_init(void)
1034{
1035 /* See RFC 2012 */
1036 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1037 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1038 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1039 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1040}
1041
6a438bbe
SH		 1042/*from STCP */
1043static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1044 tp->lost_skb_hint = NULL;
1045 tp->scoreboard_skb_hint = NULL;
1046 tp->retransmit_skb_hint = NULL;
1047 tp->forward_skb_hint = NULL;
1048 tp->fastpath_skb_hint = NULL;
1049}
1050
1da177e4
LT		 1051/* /proc */
1052enum tcp_seq_states {
1053 TCP_SEQ_STATE_LISTENING,
1054 TCP_SEQ_STATE_OPENREQ,
1055 TCP_SEQ_STATE_ESTABLISHED,
1056 TCP_SEQ_STATE_TIME_WAIT,
1057};
1058
1059struct tcp_seq_afinfo {
1060 struct module *owner;
1061 char *name;
1062 sa_family_t family;
1063 int (*seq_show) (struct seq_file *m, void *v);
1064 struct file_operations *seq_fops;
1065};
1066
1067struct tcp_iter_state {
1068 sa_family_t family;
1069 enum tcp_seq_states state;
1070 struct sock *syn_wait_sk;
1071 int bucket, sbucket, num, uid;
1072 struct seq_operations seq_ops;
1073};
1074
1075extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1076extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1077
20380731
ACM		 1078extern struct request_sock_ops tcp_request_sock_ops;
1079
1080extern int tcp_v4_destroy_sock(struct sock *sk);
1081
1082#ifdef CONFIG_PROC_FS
1083extern int tcp4_proc_init(void);
1084extern void tcp4_proc_exit(void);
1085#endif
1086
1087extern void tcp_v4_init(struct net_proto_family *ops);
1088extern void tcp_init(void);
1089
1da177e4 1090#endif /* _TCP_H */
Linux 6.x block layer and io_uring tree(s)