projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[TCP]: Access to highest_sack obsoletes forward_cnt_hint
[linux-2.6-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/list.h>
25#include <linux/tcp.h>
26#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
fb286bb2 29#include <linux/skbuff.h>
97fc2f08 30#include <linux/dmaengine.h>
cfb6eeb4 31#include <linux/crypto.h>
3f421baa
ACM		 32
33#include <net/inet_connection_sock.h>
295ff7ed 34#include <net/inet_timewait_sock.h>
77d8bf9c 35#include <net/inet_hashtables.h>
1da177e4 36#include <net/checksum.h>
2e6599cb 37#include <net/request_sock.h>
1da177e4
LT		 38#include <net/sock.h>
39#include <net/snmp.h>
40#include <net/ip.h>
c752f073
ACM		 41#include <net/tcp_states.h>
42
1da177e4
LT		 43#include <linux/seq_file.h>
44
6e04e021 45extern struct inet_hashinfo tcp_hashinfo;
1da177e4 46
1da177e4 47extern atomic_t tcp_orphan_count;
1da177e4 48extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 49
1da177e4
LT		 50#define MAX_TCP_HEADER (128 + MAX_HEADER)
51
52/*
53 * Never offer a window over 32767 without using window scaling. Some
54 * poor stacks do signed 16bit maths!
55 */
56#define MAX_TCP_WINDOW 32767U
57
58/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59#define TCP_MIN_MSS 88U
60
61/* Minimal RCV_MSS. */
62#define TCP_MIN_RCVMSS 536U
63
5d424d5a
JH		 64/* The least MTU to use for probing */
65#define TCP_BASE_MSS 512
66
1da177e4
LT		 67/* After receiving this amount of duplicate ACKs fast retransmit starts. */
68#define TCP_FASTRETRANS_THRESH 3
69
70/* Maximal reordering. */
71#define TCP_MAX_REORDERING 127
72
73/* Maximal number of ACKs sent quickly to accelerate slow-start. */
74#define TCP_MAX_QUICKACKS 16U
75
76/* urg_data states */
77#define TCP_URG_VALID 0x0100
78#define TCP_URG_NOTYET 0x0200
79#define TCP_URG_READ 0x0400
80
81#define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
86 */
87
88#define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
93 */
94
95#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
caa20d9a 96 * connection: ~180sec is RFC minimum */
1da177e4
LT		 97
98#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
caa20d9a 99 * connection: ~180sec is RFC minimum */
1da177e4
LT		 100
101
102#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
103 * socket. 7 is ~50sec-16min.
104 */
105
106
107#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
113 * TIME-WAIT timer.
114 */
115
116#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
117#if HZ >= 100
118#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119#define TCP_ATO_MIN ((unsigned)(HZ/25))
120#else
121#define TCP_DELACK_MIN 4U
122#define TCP_ATO_MIN 4U
123#endif
124#define TCP_RTO_MAX ((unsigned)(120*HZ))
125#define TCP_RTO_MIN ((unsigned)(HZ/5))
126#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
127
128#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
129 * for local resources.
130 */
131
132#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
133#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
134#define TCP_KEEPALIVE_INTVL (75*HZ)
135
136#define MAX_TCP_KEEPIDLE 32767
137#define MAX_TCP_KEEPINTVL 32767
138#define MAX_TCP_KEEPCNT 127
139#define MAX_TCP_SYNCNT 127
140
141#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
1da177e4
LT		 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 */
cfb6eeb4 165#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
1da177e4
LT		 166
167/*
168 * TCP option lengths
169 */
170
171#define TCPOLEN_MSS 4
172#define TCPOLEN_WINDOW 3
173#define TCPOLEN_SACK_PERM 2
174#define TCPOLEN_TIMESTAMP 10
cfb6eeb4 175#define TCPOLEN_MD5SIG 18
1da177e4
LT		 176
177/* But this is what stacks really send out. */
178#define TCPOLEN_TSTAMP_ALIGNED 12
179#define TCPOLEN_WSCALE_ALIGNED 4
180#define TCPOLEN_SACKPERM_ALIGNED 4
181#define TCPOLEN_SACK_BASE 2
182#define TCPOLEN_SACK_BASE_ALIGNED 4
183#define TCPOLEN_SACK_PERBLOCK 8
cfb6eeb4 184#define TCPOLEN_MD5SIG_ALIGNED 20
1da177e4 185
1da177e4
LT		 186/* Flags in tp->nonagle */
187#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
188#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 189#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 190
295ff7ed
ACM		 191extern struct inet_timewait_death_row tcp_death_row;
192
1da177e4 193/* sysctl variables for tcp */
1da177e4
LT		 194extern int sysctl_tcp_timestamps;
195extern int sysctl_tcp_window_scaling;
196extern int sysctl_tcp_sack;
197extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 198extern int sysctl_tcp_keepalive_time;
199extern int sysctl_tcp_keepalive_probes;
200extern int sysctl_tcp_keepalive_intvl;
201extern int sysctl_tcp_syn_retries;
202extern int sysctl_tcp_synack_retries;
203extern int sysctl_tcp_retries1;
204extern int sysctl_tcp_retries2;
205extern int sysctl_tcp_orphan_retries;
206extern int sysctl_tcp_syncookies;
207extern int sysctl_tcp_retrans_collapse;
208extern int sysctl_tcp_stdurg;
209extern int sysctl_tcp_rfc1337;
210extern int sysctl_tcp_abort_on_overflow;
211extern int sysctl_tcp_max_orphans;
1da177e4
LT		 212extern int sysctl_tcp_fack;
213extern int sysctl_tcp_reordering;
214extern int sysctl_tcp_ecn;
215extern int sysctl_tcp_dsack;
216extern int sysctl_tcp_mem[3];
217extern int sysctl_tcp_wmem[3];
218extern int sysctl_tcp_rmem[3];
219extern int sysctl_tcp_app_win;
220extern int sysctl_tcp_adv_win_scale;
221extern int sysctl_tcp_tw_reuse;
222extern int sysctl_tcp_frto;
3cfe3baa 223extern int sysctl_tcp_frto_response;
1da177e4 224extern int sysctl_tcp_low_latency;
95937825 225extern int sysctl_tcp_dma_copybreak;
1da177e4 226extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 227extern int sysctl_tcp_moderate_rcvbuf;
228extern int sysctl_tcp_tso_win_divisor;
9772efb9 229extern int sysctl_tcp_abc;
5d424d5a
JH		 230extern int sysctl_tcp_mtu_probing;
231extern int sysctl_tcp_base_mss;
15d99e02 232extern int sysctl_tcp_workaround_signed_windows;
35089bb2 233extern int sysctl_tcp_slow_start_after_idle;
886236c1 234extern int sysctl_tcp_max_ssthresh;
1da177e4
LT		 235
236extern atomic_t tcp_memory_allocated;
237extern atomic_t tcp_sockets_allocated;
238extern int tcp_memory_pressure;
239
1da177e4
LT		 240/*
241 * The next routines deal with comparing 32 bit unsigned ints
242 * and worry about wraparound (automatic with unsigned arithmetic).
243 */
244
245static inline int before(__u32 seq1, __u32 seq2)
246{
0d630cc0 247 return (__s32)(seq1-seq2) < 0;
1da177e4 248}
9a036b9c 249#define after(seq2, seq1) before(seq1, seq2)
1da177e4
LT		 250
251/* is s2<=s1<=s3 ? */
252static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
253{
254 return seq3 - seq2 >= seq1 - seq2;
255}
256
e4fd5da3
PE		 257static inline int tcp_too_many_orphans(struct sock *sk, int num)
258{
259 return (num > sysctl_tcp_max_orphans) ||
260 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
261 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
262}
1da177e4
LT		 263
264extern struct proto tcp_prot;
265
266DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
267#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
268#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
269#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
270#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
271#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
272#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
273
1da177e4
LT		 274extern void tcp_v4_err(struct sk_buff *skb, u32);
275
276extern void tcp_shutdown (struct sock *sk, int how);
277
278extern int tcp_v4_rcv(struct sk_buff *skb);
279
280extern int tcp_v4_remember_stamp(struct sock *sk);
281
8feaf0c0 282extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4 283
3516ffb0 284extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
1da177e4
LT		 285 struct msghdr *msg, size_t size);
286extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
287
288extern int tcp_ioctl(struct sock *sk,
289 int cmd,
290 unsigned long arg);
291
292extern int tcp_rcv_state_process(struct sock *sk,
293 struct sk_buff *skb,
294 struct tcphdr *th,
295 unsigned len);
296
297extern int tcp_rcv_established(struct sock *sk,
298 struct sk_buff *skb,
299 struct tcphdr *th,
300 unsigned len);
301
302extern void tcp_rcv_space_adjust(struct sock *sk);
303
0e4b4992
CL		 304extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
305
6d6ee43e
ACM		 306extern int tcp_twsk_unique(struct sock *sk,
307 struct sock *sktw, void *twp);
308
cfb6eeb4
YH		 309extern void tcp_twsk_destructor(struct sock *sk);
310
463c84b9
ACM		 311static inline void tcp_dec_quickack_mode(struct sock *sk,
312 const unsigned int pkts)
1da177e4 313{
463c84b9 314 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 315
463c84b9
ACM		 316 if (icsk->icsk_ack.quick) {
317 if (pkts >= icsk->icsk_ack.quick) {
318 icsk->icsk_ack.quick = 0;
fc6415bc 319 /* Leaving quickack mode we deflate ATO. */
463c84b9 320 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 321 } else
463c84b9 322 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 323 }
324}
325
463c84b9 326extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 327
1da177e4
LT		 328static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
329{
330 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
331}
332
333enum tcp_tw_status
334{
335 TCP_TW_SUCCESS = 0,
336 TCP_TW_RST = 1,
337 TCP_TW_ACK = 2,
338 TCP_TW_SYN = 3
339};
340
341
8feaf0c0 342extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 343 struct sk_buff *skb,
8feaf0c0 344 const struct tcphdr *th);
1da177e4
LT		 345
346extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 347 struct request_sock *req,
348 struct request_sock **prev);
1da177e4
LT		 349extern int tcp_child_process(struct sock *parent,
350 struct sock *child,
351 struct sk_buff *skb);
46d0de4e 352extern int tcp_use_frto(struct sock *sk);
1da177e4
LT		 353extern void tcp_enter_frto(struct sock *sk);
354extern void tcp_enter_loss(struct sock *sk, int how);
355extern void tcp_clear_retrans(struct tcp_sock *tp);
356extern void tcp_update_metrics(struct sock *sk);
357
358extern void tcp_close(struct sock *sk,
359 long timeout);
1da177e4
LT		 360extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
361
362extern int tcp_getsockopt(struct sock *sk, int level,
363 int optname,
364 char __user *optval,
365 int __user *optlen);
366extern int tcp_setsockopt(struct sock *sk, int level,
367 int optname, char __user *optval,
368 int optlen);
3fdadf7d
DM		 369extern int compat_tcp_getsockopt(struct sock *sk,
370 int level, int optname,
371 char __user *optval, int __user *optlen);
372extern int compat_tcp_setsockopt(struct sock *sk,
373 int level, int optname,
374 char __user *optval, int optlen);
1da177e4
LT		 375extern void tcp_set_keepalive(struct sock *sk, int val);
376extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
377 struct msghdr *msg,
378 size_t len, int nonblock,
379 int flags, int *addr_len);
380
1da177e4
LT		 381extern void tcp_parse_options(struct sk_buff *skb,
382 struct tcp_options_received *opt_rx,
383 int estab);
384
385/*
386 * TCP v4 functions exported for the inet6 API
387 */
388
8292a17a 389extern void tcp_v4_send_check(struct sock *sk, int len,
1da177e4
LT		 390 struct sk_buff *skb);
391
392extern int tcp_v4_conn_request(struct sock *sk,
393 struct sk_buff *skb);
394
395extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 396 struct request_sock *req,
1da177e4
LT		 397 struct sk_buff *skb);
398
399extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
400 struct sk_buff *skb,
60236fdd 401 struct request_sock *req,
1da177e4
LT		 402 struct dst_entry *dst);
403
404extern int tcp_v4_do_rcv(struct sock *sk,
405 struct sk_buff *skb);
406
407extern int tcp_v4_connect(struct sock *sk,
408 struct sockaddr *uaddr,
409 int addr_len);
410
411extern int tcp_connect(struct sock *sk);
412
413extern struct sk_buff * tcp_make_synack(struct sock *sk,
414 struct dst_entry *dst,
60236fdd 415 struct request_sock *req);
1da177e4
LT		 416
417extern int tcp_disconnect(struct sock *sk, int flags);
418
419extern void tcp_unhash(struct sock *sk);
420
1da177e4
LT		 421/* From syncookies.c */
422extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
423 struct ip_options *opt);
424extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
425 __u16 *mss);
426
427/* tcp_output.c */
428
9e412ba7
IJ		 429extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
430 int nonagle);
431extern int tcp_may_send_now(struct sock *sk);
1da177e4
LT		 432extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
433extern void tcp_xmit_retransmit_queue(struct sock *);
434extern void tcp_simple_retransmit(struct sock *);
435extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 436extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 437
438extern void tcp_send_probe0(struct sock *);
439extern void tcp_send_partial(struct sock *);
440extern int tcp_write_wakeup(struct sock *);
441extern void tcp_send_fin(struct sock *sk);
dd0fc66f 442extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
1da177e4 443extern int tcp_send_synack(struct sock *);
c1b4a7e6 444extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 445extern void tcp_send_ack(struct sock *sk);
446extern void tcp_send_delayed_ack(struct sock *sk);
447
a762a980
DM		 448/* tcp_input.c */
449extern void tcp_cwnd_application_limited(struct sock *sk);
450
1da177e4
LT		 451/* tcp_timer.c */
452extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 453static inline void tcp_clear_xmit_timers(struct sock *sk)
454{
455 inet_csk_clear_xmit_timers(sk);
456}
1da177e4 457
1da177e4
LT		 458extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
459extern unsigned int tcp_current_mss(struct sock *sk, int large);
460
17b085ea 461/* tcp.c */
1da177e4
LT		 462extern void tcp_get_info(struct sock *, struct tcp_info *);
463
464/* Read 'sendfile()'-style from a TCP socket */
465typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
466 unsigned int, size_t);
467extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
468 sk_read_actor_t recv_actor);
469
40efc6fa 470extern void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 471
5d424d5a
JH		 472extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
473extern int tcp_mss_to_mtu(struct sock *sk, int mss);
474extern void tcp_mtup_init(struct sock *sk);
475
40efc6fa 476static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 477{
478 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
479 ntohl(TCP_FLAG_ACK) |
480 snd_wnd);
481}
482
40efc6fa 483static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 484{
485 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
486}
487
9e412ba7 488static inline void tcp_fast_path_check(struct sock *sk)
1da177e4 489{
9e412ba7
IJ		 490 struct tcp_sock *tp = tcp_sk(sk);
491
b03efcfb 492 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 493 tp->rcv_wnd &&
494 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
495 !tp->urg_data)
496 tcp_fast_path_on(tp);
497}
498
499/* Compute the actual receive window we are currently advertising.
500 * Rcv_nxt can be after the window if our peer push more data
501 * than the offered window.
502 */
40efc6fa 503static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 504{
505 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
506
507 if (win < 0)
508 win = 0;
509 return (u32) win;
510}
511
512/* Choose a new window, without checks for shrinking, and without
513 * scaling applied to the result. The caller does these things
514 * if necessary. This is a "raw" window selection.
515 */
516extern u32 __tcp_select_window(struct sock *sk);
517
518/* TCP timestamps are only 32-bits, this causes a slight
519 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 520 * of jiffies in the buffer control blocks below. We decided
521 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 522 * casts with the following macro.
523 */
524#define tcp_time_stamp ((__u32)(jiffies))
525
caa20d9a 526/* This is what the send packet queuing engine uses to pass
1da177e4
LT		 527 * TCP per-packet control information to the transmission
528 * code. We also store the host-order sequence numbers in
529 * here too. This is 36 bytes on 32-bit architectures,
530 * 40 bytes on 64-bit machines, if this grows please adjust
531 * skbuff.h:skbuff->cb[xxx] size appropriately.
532 */
533struct tcp_skb_cb {
534 union {
535 struct inet_skb_parm h4;
536#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
537 struct inet6_skb_parm h6;
538#endif
539 } header; /* For incoming frames */
540 __u32 seq; /* Starting sequence number */
541 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
542 __u32 when; /* used to compute rtt's */
543 __u8 flags; /* TCP header flags. */
544
545 /* NOTE: These must match up to the flags byte in a
546 * real TCP header.
547 */
548#define TCPCB_FLAG_FIN 0x01
549#define TCPCB_FLAG_SYN 0x02
550#define TCPCB_FLAG_RST 0x04
551#define TCPCB_FLAG_PSH 0x08
552#define TCPCB_FLAG_ACK 0x10
553#define TCPCB_FLAG_URG 0x20
554#define TCPCB_FLAG_ECE 0x40
555#define TCPCB_FLAG_CWR 0x80
556
557 __u8 sacked; /* State flags for SACK/FACK. */
558#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
559#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
560#define TCPCB_LOST 0x04 /* SKB is lost */
561#define TCPCB_TAGBITS 0x07 /* All tag bits */
562
563#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
564#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
565
caa20d9a 566#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
1da177e4
LT		 567
568#define TCPCB_AT_TAIL (TCPCB_URG)
569
570 __u16 urg_ptr; /* Valid w/URG flags is set. */
571 __u32 ack_seq; /* Sequence number ACK'd */
572};
573
574#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
575
576#include <net/tcp_ecn.h>
577
578/* Due to TSO, an SKB can be composed of multiple actual
579 * packets. To keep these tracked properly, we use this.
580 */
581static inline int tcp_skb_pcount(const struct sk_buff *skb)
582{
7967168c 583 return skb_shinfo(skb)->gso_segs;
1da177e4
LT		 584}
585
586/* This is valid iff tcp_skb_pcount() > 1. */
587static inline int tcp_skb_mss(const struct sk_buff *skb)
588{
7967168c 589 return skb_shinfo(skb)->gso_size;
1da177e4
LT		 590}
591
592static inline void tcp_dec_pcount_approx(__u32 *count,
593 const struct sk_buff *skb)
594{
595 if (*count) {
596 *count -= tcp_skb_pcount(skb);
597 if ((int)*count < 0)
598 *count = 0;
599 }
600}
601
9e412ba7 602static inline void tcp_packets_out_inc(struct sock *sk,
1da177e4
LT		 603 const struct sk_buff *skb)
604{
9e412ba7 605 struct tcp_sock *tp = tcp_sk(sk);
1da177e4
LT		 606 int orig = tp->packets_out;
607
608 tp->packets_out += tcp_skb_pcount(skb);
609 if (!orig)
3f421baa
ACM		 610 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
611 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 612}
613
614static inline void tcp_packets_out_dec(struct tcp_sock *tp,
615 const struct sk_buff *skb)
616{
617 tp->packets_out -= tcp_skb_pcount(skb);
618}
619
317a76f9
SH		 620/* Events passed to congestion control interface */
621enum tcp_ca_event {
622 CA_EVENT_TX_START, /* first transmit when no packets in flight */
623 CA_EVENT_CWND_RESTART, /* congestion window restart */
624 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
625 CA_EVENT_FRTO, /* fast recovery timeout */
626 CA_EVENT_LOSS, /* loss timeout */
627 CA_EVENT_FAST_ACK, /* in sequence ack */
628 CA_EVENT_SLOW_ACK, /* other ack */
629};
630
631/*
632 * Interface for adding new TCP congestion control handlers
633 */
634#define TCP_CA_NAME_MAX 16
3ff825b2
SH		 635#define TCP_CA_MAX 128
636#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
637
164891aa
SH		 638#define TCP_CONG_NON_RESTRICTED 0x1
639#define TCP_CONG_RTT_STAMP 0x2
640
317a76f9
SH		 641struct tcp_congestion_ops {
642 struct list_head list;
164891aa 643 unsigned long flags;
317a76f9
SH		 644
645 /* initialize private data (optional) */
6687e988 646 void (*init)(struct sock *sk);
317a76f9 647 /* cleanup private data (optional) */
6687e988 648 void (*release)(struct sock *sk);
317a76f9
SH		 649
650 /* return slow start threshold (required) */
6687e988 651 u32 (*ssthresh)(struct sock *sk);
317a76f9 652 /* lower bound for congestion window (optional) */
72dc5b92 653 u32 (*min_cwnd)(const struct sock *sk);
317a76f9 654 /* do new cwnd calculation (required) */
16751347 655 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight, int good_ack);
317a76f9 656 /* call before changing ca_state (optional) */
6687e988 657 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 658 /* call when cwnd event occurs (optional) */
6687e988 659 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 660 /* new value of cwnd after loss (optional) */
6687e988 661 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 662 /* hook for packet ack accounting (optional) */
30cfd0ba 663 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
73c1f4a0 664 /* get info for inet_diag (optional) */
6687e988 665 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 666
667 char name[TCP_CA_NAME_MAX];
668 struct module *owner;
669};
670
671extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
672extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
673
6687e988
ACM		 674extern void tcp_init_congestion_control(struct sock *sk);
675extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 676extern int tcp_set_default_congestion_control(const char *name);
677extern void tcp_get_default_congestion_control(char *name);
3ff825b2 678extern void tcp_get_available_congestion_control(char *buf, size_t len);
ce7bc3bf
SH		 679extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
680extern int tcp_set_allowed_congestion_control(char *allowed);
6687e988 681extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 682extern void tcp_slow_start(struct tcp_sock *tp);
317a76f9 683
5f8ef48d 684extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988 685extern u32 tcp_reno_ssthresh(struct sock *sk);
16751347 686extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight, int flag);
72dc5b92 687extern u32 tcp_reno_min_cwnd(const struct sock *sk);
a8acfbac 688extern struct tcp_congestion_ops tcp_reno;
317a76f9 689
6687e988 690static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 691{
6687e988
ACM		 692 struct inet_connection_sock *icsk = inet_csk(sk);
693
694 if (icsk->icsk_ca_ops->set_state)
695 icsk->icsk_ca_ops->set_state(sk, ca_state);
696 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 697}
698
6687e988 699static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 700{
6687e988
ACM		 701 const struct inet_connection_sock *icsk = inet_csk(sk);
702
703 if (icsk->icsk_ca_ops->cwnd_event)
704 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 705}
706
1da177e4
LT		 707/* This determines how many packets are "in the network" to the best
708 * of our knowledge. In many cases it is conservative, but where
709 * detailed information is available from the receiver (via SACK
710 * blocks etc.) we can make more aggressive calculations.
711 *
712 * Use this for decisions involving congestion control, use just
713 * tp->packets_out to determine if the send queue is empty or not.
714 *
715 * Read this equation as:
716 *
717 * "Packets sent once on transmission queue" MINUS
718 * "Packets left network, but not honestly ACKed yet" PLUS
719 * "Packets fast retransmitted"
720 */
40efc6fa 721static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4
LT		 722{
723 return (tp->packets_out - tp->left_out + tp->retrans_out);
724}
725
1da177e4
LT		 726/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
727 * The exception is rate halving phase, when cwnd is decreasing towards
728 * ssthresh.
729 */
6687e988 730static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 731{
6687e988
ACM		 732 const struct tcp_sock *tp = tcp_sk(sk);
733 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 734 return tp->snd_ssthresh;
735 else
736 return max(tp->snd_ssthresh,
737 ((tp->snd_cwnd >> 1) +
738 (tp->snd_cwnd >> 2)));
739}
740
741static inline void tcp_sync_left_out(struct tcp_sock *tp)
742{
0ec96822
IJ		 743 BUG_ON(tp->rx_opt.sack_ok &&
744 (tp->sacked_out + tp->lost_out > tp->packets_out));
1da177e4
LT		 745 tp->left_out = tp->sacked_out + tp->lost_out;
746}
747
3cfe3baa 748extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
1da177e4
LT		 749extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
750
751/* Slow start with delack produces 3 packets of burst, so that
752 * it is safe "de facto".
753 */
754static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
755{
756 return 3;
757}
758
f4805ede
SH		 759/* RFC2861 Check whether we are limited by application or congestion window
760 * This is the inverse of cwnd check in tcp_tso_should_defer
761 */
762static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
763{
764 const struct tcp_sock *tp = tcp_sk(sk);
765 u32 left;
766
767 if (in_flight >= tp->snd_cwnd)
768 return 1;
769
bcd76111 770 if (!sk_can_gso(sk))
f4805ede
SH		 771 return 0;
772
773 left = tp->snd_cwnd - in_flight;
774 if (sysctl_tcp_tso_win_divisor)
775 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
776 else
777 return left <= tcp_max_burst(tp);
778}
779
40efc6fa
SH		 780static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
781 const struct sk_buff *skb)
1da177e4
LT		 782{
783 if (skb->len < mss)
784 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
785}
786
9e412ba7 787static inline void tcp_check_probe_timer(struct sock *sk)
1da177e4 788{
9e412ba7 789 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 790 const struct inet_connection_sock *icsk = inet_csk(sk);
9e412ba7 791
463c84b9 792 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 793 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
794 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 795}
796
9e412ba7 797static inline void tcp_push_pending_frames(struct sock *sk)
1da177e4 798{
9e412ba7
IJ		 799 struct tcp_sock *tp = tcp_sk(sk);
800
801 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
1da177e4
LT		 802}
803
40efc6fa 804static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 805{
806 tp->snd_wl1 = seq;
807}
808
40efc6fa 809static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 810{
811 tp->snd_wl1 = seq;
812}
813
1da177e4
LT		 814/*
815 * Calculate(/check) TCP checksum
816 */
ba7808ea
FD		 817static inline __sum16 tcp_v4_check(int len, __be32 saddr,
818 __be32 daddr, __wsum base)
1da177e4
LT		 819{
820 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
821}
822
b51655b9 823static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 824{
fb286bb2 825 return __skb_checksum_complete(skb);
1da177e4
LT		 826}
827
40efc6fa 828static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4 829{
60476372 830 return !skb_csum_unnecessary(skb) &&
1da177e4
LT		 831 __tcp_checksum_complete(skb);
832}
833
834/* Prequeue for VJ style copy to user, combined with checksumming. */
835
40efc6fa 836static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 837{
838 tp->ucopy.task = NULL;
839 tp->ucopy.len = 0;
840 tp->ucopy.memory = 0;
841 skb_queue_head_init(&tp->ucopy.prequeue);
97fc2f08
CL		 842#ifdef CONFIG_NET_DMA
843 tp->ucopy.dma_chan = NULL;
844 tp->ucopy.wakeup = 0;
845 tp->ucopy.pinned_list = NULL;
846 tp->ucopy.dma_cookie = 0;
847#endif
1da177e4
LT		 848}
849
850/* Packet is added to VJ-style prequeue for processing in process
851 * context, if a reader task is waiting. Apparently, this exciting
852 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
853 * failed somewhere. Latency? Burstiness? Well, at least now we will
854 * see, why it failed. 8)8) --ANK
855 *
856 * NOTE: is this not too big to inline?
857 */
40efc6fa 858static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 859{
860 struct tcp_sock *tp = tcp_sk(sk);
861
862 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
863 __skb_queue_tail(&tp->ucopy.prequeue, skb);
864 tp->ucopy.memory += skb->truesize;
865 if (tp->ucopy.memory > sk->sk_rcvbuf) {
866 struct sk_buff *skb1;
867
868 BUG_ON(sock_owned_by_user(sk));
869
870 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
871 sk->sk_backlog_rcv(sk, skb1);
872 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
873 }
874
875 tp->ucopy.memory = 0;
876 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
877 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 878 if (!inet_csk_ack_scheduled(sk))
879 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
3f421baa
ACM		 880 (3 * TCP_RTO_MIN) / 4,
881 TCP_RTO_MAX);
1da177e4
LT		 882 }
883 return 1;
884 }
885 return 0;
886}
887
888
889#undef STATE_TRACE
890
891#ifdef STATE_TRACE
892static const char *statename[]={
893 "Unused","Established","Syn Sent","Syn Recv",
894 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
895 "Close Wait","Last ACK","Listen","Closing"
896};
897#endif
898
40efc6fa 899static inline void tcp_set_state(struct sock *sk, int state)
1da177e4
LT		 900{
901 int oldstate = sk->sk_state;
902
903 switch (state) {
904 case TCP_ESTABLISHED:
905 if (oldstate != TCP_ESTABLISHED)
906 TCP_INC_STATS(TCP_MIB_CURRESTAB);
907 break;
908
909 case TCP_CLOSE:
910 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
911 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
912
913 sk->sk_prot->unhash(sk);
463c84b9 914 if (inet_csk(sk)->icsk_bind_hash &&
1da177e4 915 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2d8c4ce5 916 inet_put_port(&tcp_hashinfo, sk);
1da177e4
LT		 917 /* fall through */
918 default:
919 if (oldstate==TCP_ESTABLISHED)
920 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
921 }
922
923 /* Change state AFTER socket is unhashed to avoid closed
924 * socket sitting in hash tables.
925 */
926 sk->sk_state = state;
927
928#ifdef STATE_TRACE
929 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
930#endif
931}
932
4ac02bab 933extern void tcp_done(struct sock *sk);
1da177e4 934
40efc6fa 935static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 936{
937 rx_opt->dsack = 0;
938 rx_opt->eff_sacks = 0;
939 rx_opt->num_sacks = 0;
940}
941
1da177e4
LT		 942/* Determine a window scaling and initial window to offer. */
943extern void tcp_select_initial_window(int __space, __u32 mss,
944 __u32 *rcv_wnd, __u32 *window_clamp,
945 int wscale_ok, __u8 *rcv_wscale);
946
947static inline int tcp_win_from_space(int space)
948{
949 return sysctl_tcp_adv_win_scale<=0 ?
950 (space>>(-sysctl_tcp_adv_win_scale)) :
951 space - (space>>sysctl_tcp_adv_win_scale);
952}
953
954/* Note: caller must be prepared to deal with negative returns */
955static inline int tcp_space(const struct sock *sk)
956{
957 return tcp_win_from_space(sk->sk_rcvbuf -
958 atomic_read(&sk->sk_rmem_alloc));
959}
960
961static inline int tcp_full_space(const struct sock *sk)
962{
963 return tcp_win_from_space(sk->sk_rcvbuf);
964}
965
40efc6fa
SH		 966static inline void tcp_openreq_init(struct request_sock *req,
967 struct tcp_options_received *rx_opt,
968 struct sk_buff *skb)
1da177e4 969{
2e6599cb
ACM		 970 struct inet_request_sock *ireq = inet_rsk(req);
971
1da177e4 972 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 973 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 974 req->mss = rx_opt->mss_clamp;
975 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 976 ireq->tstamp_ok = rx_opt->tstamp_ok;
977 ireq->sack_ok = rx_opt->sack_ok;
978 ireq->snd_wscale = rx_opt->snd_wscale;
979 ireq->wscale_ok = rx_opt->wscale_ok;
980 ireq->acked = 0;
981 ireq->ecn_ok = 0;
aa8223c7 982 ireq->rmt_port = tcp_hdr(skb)->source;
1da177e4
LT		 983}
984
985extern void tcp_enter_memory_pressure(void);
986
1da177e4
LT		 987static inline int keepalive_intvl_when(const struct tcp_sock *tp)
988{
989 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
990}
991
992static inline int keepalive_time_when(const struct tcp_sock *tp)
993{
994 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
995}
996
463c84b9 997static inline int tcp_fin_time(const struct sock *sk)
1da177e4 998{
463c84b9
ACM		 999 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1000 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 1001
463c84b9
ACM		 1002 if (fin_timeout < (rto << 2) - (rto >> 1))
1003 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 1004
1005 return fin_timeout;
1006}
1007
1008static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1009{
1010 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1011 return 0;
9d729f72 1012 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1da177e4
LT		 1013 return 0;
1014
1015 /* RST segments are not recommended to carry timestamp,
1016 and, if they do, it is recommended to ignore PAWS because
1017 "their cleanup function should take precedence over timestamps."
1018 Certainly, it is mistake. It is necessary to understand the reasons
1019 of this constraint to relax it: if peer reboots, clock may go
1020 out-of-sync and half-open connections will not be reset.
1021 Actually, the problem would be not existing if all
1022 the implementations followed draft about maintaining clock
1023 via reboots. Linux-2.2 DOES NOT!
1024
1025 However, we can relax time bounds for RST segments to MSL.
1026 */
9d729f72 1027 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1da177e4
LT		 1028 return 0;
1029 return 1;
1030}
1031
1da177e4
LT		 1032#define TCP_CHECK_TIMER(sk) do { } while (0)
1033
1da177e4
LT		 1034static inline void tcp_mib_init(void)
1035{
1036 /* See RFC 2012 */
1037 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1038 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1039 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1040 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1041}
1042
6a438bbe
SH		 1043/*from STCP */
1044static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1045 tp->lost_skb_hint = NULL;
1046 tp->scoreboard_skb_hint = NULL;
1047 tp->retransmit_skb_hint = NULL;
1048 tp->forward_skb_hint = NULL;
1049 tp->fastpath_skb_hint = NULL;
1050}
1051
cfb6eeb4
YH		 1052/* MD5 Signature */
1053struct crypto_hash;
1054
1055/* - key database */
1056struct tcp_md5sig_key {
1057 u8 *key;
1058 u8 keylen;
1059};
1060
1061struct tcp4_md5sig_key {
f8ab18d2 1062 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1063 __be32 addr;
1064};
1065
1066struct tcp6_md5sig_key {
f8ab18d2 1067 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1068#if 0
1069 u32 scope_id; /* XXX */
1070#endif
1071 struct in6_addr addr;
1072};
1073
1074/* - sock block */
1075struct tcp_md5sig_info {
1076 struct tcp4_md5sig_key *keys4;
1077#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1078 struct tcp6_md5sig_key *keys6;
1079 u32 entries6;
1080 u32 alloced6;
1081#endif
1082 u32 entries4;
1083 u32 alloced4;
1084};
1085
1086/* - pseudo header */
1087struct tcp4_pseudohdr {
1088 __be32 saddr;
1089 __be32 daddr;
1090 __u8 pad;
1091 __u8 protocol;
1092 __be16 len;
1093};
1094
1095struct tcp6_pseudohdr {
1096 struct in6_addr saddr;
1097 struct in6_addr daddr;
1098 __be32 len;
1099 __be32 protocol; /* including padding */
1100};
1101
1102union tcp_md5sum_block {
1103 struct tcp4_pseudohdr ip4;
1104#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1105 struct tcp6_pseudohdr ip6;
1106#endif
1107};
1108
1109/* - pool: digest algorithm, hash description and scratch buffer */
1110struct tcp_md5sig_pool {
1111 struct hash_desc md5_desc;
1112 union tcp_md5sum_block md5_blk;
1113};
1114
1115#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1116
1117/* - functions */
1118extern int tcp_v4_calc_md5_hash(char *md5_hash,
1119 struct tcp_md5sig_key *key,
1120 struct sock *sk,
1121 struct dst_entry *dst,
1122 struct request_sock *req,
1123 struct tcphdr *th,
1124 int protocol, int tcplen);
1125extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1126 struct sock *addr_sk);
1127
1128extern int tcp_v4_md5_do_add(struct sock *sk,
1129 __be32 addr,
1130 u8 *newkey,
1131 u8 newkeylen);
1132
1133extern int tcp_v4_md5_do_del(struct sock *sk,
8e5200f5 1134 __be32 addr);
cfb6eeb4
YH		 1135
1136extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1137extern void tcp_free_md5sig_pool(void);
1138
1139extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1140extern void __tcp_put_md5sig_pool(void);
1141
1142static inline
1143struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1144{
1145 int cpu = get_cpu();
1146 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1147 if (!ret)
1148 put_cpu();
1149 return ret;
1150}
1151
1152static inline void tcp_put_md5sig_pool(void)
1153{
1154 __tcp_put_md5sig_pool();
1155 put_cpu();
1156}
1157
fe067e8a
DM		 1158/* write queue abstraction */
1159static inline void tcp_write_queue_purge(struct sock *sk)
1160{
1161 struct sk_buff *skb;
1162
1163 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1164 sk_stream_free_skb(sk, skb);
1165 sk_stream_mem_reclaim(sk);
1166}
1167
1168static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1169{
1170 struct sk_buff *skb = sk->sk_write_queue.next;
1171 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1172 return NULL;
1173 return skb;
1174}
1175
1176static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1177{
1178 struct sk_buff *skb = sk->sk_write_queue.prev;
1179 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1180 return NULL;
1181 return skb;
1182}
1183
1184static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1185{
1186 return skb->next;
1187}
1188
1189#define tcp_for_write_queue(skb, sk) \
1190 for (skb = (sk)->sk_write_queue.next; \
1191 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1192 skb = skb->next)
1193
1194#define tcp_for_write_queue_from(skb, sk) \
1195 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1196 skb = skb->next)
1197
1198static inline struct sk_buff *tcp_send_head(struct sock *sk)
1199{
1200 return sk->sk_send_head;
1201}
1202
1203static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1204{
d551e454
IJ		 1205 struct tcp_sock *tp = tcp_sk(sk);
1206
fe067e8a
DM		 1207 sk->sk_send_head = skb->next;
1208 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1209 sk->sk_send_head = NULL;
d551e454
IJ		 1210 /* Don't override Nagle indefinately with F-RTO */
1211 if (tp->frto_counter == 2)
1212 tp->frto_counter = 3;
fe067e8a
DM		 1213}
1214
1215static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1216{
1217 if (sk->sk_send_head == skb_unlinked)
1218 sk->sk_send_head = NULL;
1219}
1220
1221static inline void tcp_init_send_head(struct sock *sk)
1222{
1223 sk->sk_send_head = NULL;
1224}
1225
1226static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1227{
1228 __skb_queue_tail(&sk->sk_write_queue, skb);
1229}
1230
1231static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1232{
1233 __tcp_add_write_queue_tail(sk, skb);
1234
1235 /* Queue it, remembering where we must start sending. */
1236 if (sk->sk_send_head == NULL)
1237 sk->sk_send_head = skb;
1238}
1239
1240static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1241{
1242 __skb_queue_head(&sk->sk_write_queue, skb);
1243}
1244
1245/* Insert buff after skb on the write queue of sk. */
1246static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1247 struct sk_buff *buff,
1248 struct sock *sk)
1249{
1250 __skb_append(skb, buff, &sk->sk_write_queue);
1251}
1252
1253/* Insert skb between prev and next on the write queue of sk. */
1254static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1255 struct sk_buff *skb,
1256 struct sock *sk)
1257{
1258 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1259}
1260
1261static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1262{
1263 __skb_unlink(skb, &sk->sk_write_queue);
1264}
1265
1266static inline int tcp_skb_is_last(const struct sock *sk,
1267 const struct sk_buff *skb)
1268{
1269 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1270}
1271
1272static inline int tcp_write_queue_empty(struct sock *sk)
1273{
1274 return skb_queue_empty(&sk->sk_write_queue);
1275}
1276
1da177e4
LT		 1277/* /proc */
1278enum tcp_seq_states {
1279 TCP_SEQ_STATE_LISTENING,
1280 TCP_SEQ_STATE_OPENREQ,
1281 TCP_SEQ_STATE_ESTABLISHED,
1282 TCP_SEQ_STATE_TIME_WAIT,
1283};
1284
1285struct tcp_seq_afinfo {
1286 struct module *owner;
1287 char *name;
1288 sa_family_t family;
1289 int (*seq_show) (struct seq_file *m, void *v);
1290 struct file_operations *seq_fops;
1291};
1292
1293struct tcp_iter_state {
1294 sa_family_t family;
1295 enum tcp_seq_states state;
1296 struct sock *syn_wait_sk;
1297 int bucket, sbucket, num, uid;
1298 struct seq_operations seq_ops;
1299};
1300
1301extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1302extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1303
20380731
ACM		 1304extern struct request_sock_ops tcp_request_sock_ops;
1305
1306extern int tcp_v4_destroy_sock(struct sock *sk);
1307
a430a43d 1308extern int tcp_v4_gso_send_check(struct sk_buff *skb);
576a30eb 1309extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
f4c50d99 1310
20380731
ACM		 1311#ifdef CONFIG_PROC_FS
1312extern int tcp4_proc_init(void);
1313extern void tcp4_proc_exit(void);
1314#endif
1315
cfb6eeb4
YH		 1316/* TCP af-specific functions */
1317struct tcp_sock_af_ops {
1318#ifdef CONFIG_TCP_MD5SIG
1319 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1320 struct sock *addr_sk);
1321 int (*calc_md5_hash) (char *location,
1322 struct tcp_md5sig_key *md5,
1323 struct sock *sk,
1324 struct dst_entry *dst,
1325 struct request_sock *req,
1326 struct tcphdr *th,
1327 int protocol, int len);
1328 int (*md5_add) (struct sock *sk,
1329 struct sock *addr_sk,
1330 u8 *newkey,
1331 u8 len);
1332 int (*md5_parse) (struct sock *sk,
1333 char __user *optval,
1334 int optlen);
1335#endif
1336};
1337
1338struct tcp_request_sock_ops {
1339#ifdef CONFIG_TCP_MD5SIG
1340 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1341 struct request_sock *req);
1342#endif
1343};
1344
20380731
ACM		 1345extern void tcp_v4_init(struct net_proto_family *ops);
1346extern void tcp_init(void);
1347
1da177e4 1348#endif /* _TCP_H */
Linux 6.x block layer and io_uring tree(s)