projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[TCP]: Reduce sacked_out with reno when purging write_queue
[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 41#include <net/tcp_states.h>
bdf1ee5d 42#include <net/inet_ecn.h>
c752f073 43
1da177e4
LT		 44#include <linux/seq_file.h>
45
6e04e021 46extern struct inet_hashinfo tcp_hashinfo;
1da177e4 47
1da177e4 48extern atomic_t tcp_orphan_count;
1da177e4 49extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 50
1da177e4
LT		 51#define MAX_TCP_HEADER (128 + MAX_HEADER)
52
53/*
54 * Never offer a window over 32767 without using window scaling. Some
55 * poor stacks do signed 16bit maths!
56 */
57#define MAX_TCP_WINDOW 32767U
58
59/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
60#define TCP_MIN_MSS 88U
61
62/* Minimal RCV_MSS. */
63#define TCP_MIN_RCVMSS 536U
64
5d424d5a
JH		 65/* The least MTU to use for probing */
66#define TCP_BASE_MSS 512
67
1da177e4
LT		 68/* After receiving this amount of duplicate ACKs fast retransmit starts. */
69#define TCP_FASTRETRANS_THRESH 3
70
71/* Maximal reordering. */
72#define TCP_MAX_REORDERING 127
73
74/* Maximal number of ACKs sent quickly to accelerate slow-start. */
75#define TCP_MAX_QUICKACKS 16U
76
77/* urg_data states */
78#define TCP_URG_VALID 0x0100
79#define TCP_URG_NOTYET 0x0200
80#define TCP_URG_READ 0x0400
81
82#define TCP_RETR1 3 /*
83 * This is how many retries it does before it
84 * tries to figure out if the gateway is
85 * down. Minimal RFC value is 3; it corresponds
86 * to ~3sec-8min depending on RTO.
87 */
88
89#define TCP_RETR2 15 /*
90 * This should take at least
91 * 90 minutes to time out.
92 * RFC1122 says that the limit is 100 sec.
93 * 15 is ~13-30min depending on RTO.
94 */
95
96#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
caa20d9a 97 * connection: ~180sec is RFC minimum */
1da177e4
LT		 98
99#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
caa20d9a 100 * connection: ~180sec is RFC minimum */
1da177e4
LT		 101
102
103#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
104 * socket. 7 is ~50sec-16min.
105 */
106
107
108#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
109 * state, about 60 seconds */
110#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
111 /* BSD style FIN_WAIT2 deadlock breaker.
112 * It used to be 3min, new value is 60sec,
113 * to combine FIN-WAIT-2 timeout with
114 * TIME-WAIT timer.
115 */
116
117#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118#if HZ >= 100
119#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
120#define TCP_ATO_MIN ((unsigned)(HZ/25))
121#else
122#define TCP_DELACK_MIN 4U
123#define TCP_ATO_MIN 4U
124#endif
125#define TCP_RTO_MAX ((unsigned)(120*HZ))
126#define TCP_RTO_MIN ((unsigned)(HZ/5))
127#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
128
129#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
130 * for local resources.
131 */
132
133#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
134#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
135#define TCP_KEEPALIVE_INTVL (75*HZ)
136
137#define MAX_TCP_KEEPIDLE 32767
138#define MAX_TCP_KEEPINTVL 32767
139#define MAX_TCP_KEEPCNT 127
140#define MAX_TCP_SYNCNT 127
141
142#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
1da177e4
LT		 143
144#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
145#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
146 * after this time. It should be equal
147 * (or greater than) TCP_TIMEWAIT_LEN
148 * to provide reliability equal to one
149 * provided by timewait state.
150 */
151#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
152 * timestamps. It must be less than
153 * minimal timewait lifetime.
154 */
1da177e4
LT		 155/*
156 * TCP option
157 */
158
159#define TCPOPT_NOP 1 /* Padding */
160#define TCPOPT_EOL 0 /* End of options */
161#define TCPOPT_MSS 2 /* Segment size negotiating */
162#define TCPOPT_WINDOW 3 /* Window scaling */
163#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
164#define TCPOPT_SACK 5 /* SACK Block */
165#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
cfb6eeb4 166#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
1da177e4
LT		 167
168/*
169 * TCP option lengths
170 */
171
172#define TCPOLEN_MSS 4
173#define TCPOLEN_WINDOW 3
174#define TCPOLEN_SACK_PERM 2
175#define TCPOLEN_TIMESTAMP 10
cfb6eeb4 176#define TCPOLEN_MD5SIG 18
1da177e4
LT		 177
178/* But this is what stacks really send out. */
179#define TCPOLEN_TSTAMP_ALIGNED 12
180#define TCPOLEN_WSCALE_ALIGNED 4
181#define TCPOLEN_SACKPERM_ALIGNED 4
182#define TCPOLEN_SACK_BASE 2
183#define TCPOLEN_SACK_BASE_ALIGNED 4
184#define TCPOLEN_SACK_PERBLOCK 8
cfb6eeb4 185#define TCPOLEN_MD5SIG_ALIGNED 20
1da177e4 186
1da177e4
LT		 187/* Flags in tp->nonagle */
188#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
189#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 190#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 191
295ff7ed
ACM		 192extern struct inet_timewait_death_row tcp_death_row;
193
1da177e4 194/* sysctl variables for tcp */
1da177e4
LT		 195extern int sysctl_tcp_timestamps;
196extern int sysctl_tcp_window_scaling;
197extern int sysctl_tcp_sack;
198extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 199extern int sysctl_tcp_keepalive_time;
200extern int sysctl_tcp_keepalive_probes;
201extern int sysctl_tcp_keepalive_intvl;
202extern int sysctl_tcp_syn_retries;
203extern int sysctl_tcp_synack_retries;
204extern int sysctl_tcp_retries1;
205extern int sysctl_tcp_retries2;
206extern int sysctl_tcp_orphan_retries;
207extern int sysctl_tcp_syncookies;
208extern int sysctl_tcp_retrans_collapse;
209extern int sysctl_tcp_stdurg;
210extern int sysctl_tcp_rfc1337;
211extern int sysctl_tcp_abort_on_overflow;
212extern int sysctl_tcp_max_orphans;
1da177e4
LT		 213extern int sysctl_tcp_fack;
214extern int sysctl_tcp_reordering;
215extern int sysctl_tcp_ecn;
216extern int sysctl_tcp_dsack;
217extern int sysctl_tcp_mem[3];
218extern int sysctl_tcp_wmem[3];
219extern int sysctl_tcp_rmem[3];
220extern int sysctl_tcp_app_win;
221extern int sysctl_tcp_adv_win_scale;
222extern int sysctl_tcp_tw_reuse;
223extern int sysctl_tcp_frto;
3cfe3baa 224extern int sysctl_tcp_frto_response;
1da177e4 225extern int sysctl_tcp_low_latency;
95937825 226extern int sysctl_tcp_dma_copybreak;
1da177e4 227extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 228extern int sysctl_tcp_moderate_rcvbuf;
229extern int sysctl_tcp_tso_win_divisor;
9772efb9 230extern int sysctl_tcp_abc;
5d424d5a
JH		 231extern int sysctl_tcp_mtu_probing;
232extern int sysctl_tcp_base_mss;
15d99e02 233extern int sysctl_tcp_workaround_signed_windows;
35089bb2 234extern int sysctl_tcp_slow_start_after_idle;
886236c1 235extern int sysctl_tcp_max_ssthresh;
1da177e4
LT		 236
237extern atomic_t tcp_memory_allocated;
238extern atomic_t tcp_sockets_allocated;
239extern int tcp_memory_pressure;
240
1da177e4
LT		 241/*
242 * The next routines deal with comparing 32 bit unsigned ints
243 * and worry about wraparound (automatic with unsigned arithmetic).
244 */
245
246static inline int before(__u32 seq1, __u32 seq2)
247{
0d630cc0 248 return (__s32)(seq1-seq2) < 0;
1da177e4 249}
9a036b9c 250#define after(seq2, seq1) before(seq1, seq2)
1da177e4
LT		 251
252/* is s2<=s1<=s3 ? */
253static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
254{
255 return seq3 - seq2 >= seq1 - seq2;
256}
257
e4fd5da3
PE		 258static inline int tcp_too_many_orphans(struct sock *sk, int num)
259{
260 return (num > sysctl_tcp_max_orphans) ||
261 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
262 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
263}
1da177e4
LT		 264
265extern struct proto tcp_prot;
266
267DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
268#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
269#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
270#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
271#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
272#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
273#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
274
1da177e4
LT		 275extern void tcp_v4_err(struct sk_buff *skb, u32);
276
277extern void tcp_shutdown (struct sock *sk, int how);
278
279extern int tcp_v4_rcv(struct sk_buff *skb);
280
281extern int tcp_v4_remember_stamp(struct sock *sk);
282
8feaf0c0 283extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4 284
3516ffb0 285extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
1da177e4
LT		 286 struct msghdr *msg, size_t size);
287extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
288
289extern int tcp_ioctl(struct sock *sk,
290 int cmd,
291 unsigned long arg);
292
293extern int tcp_rcv_state_process(struct sock *sk,
294 struct sk_buff *skb,
295 struct tcphdr *th,
296 unsigned len);
297
298extern int tcp_rcv_established(struct sock *sk,
299 struct sk_buff *skb,
300 struct tcphdr *th,
301 unsigned len);
302
303extern void tcp_rcv_space_adjust(struct sock *sk);
304
0e4b4992
CL		 305extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
306
6d6ee43e
ACM		 307extern int tcp_twsk_unique(struct sock *sk,
308 struct sock *sktw, void *twp);
309
cfb6eeb4
YH		 310extern void tcp_twsk_destructor(struct sock *sk);
311
463c84b9
ACM		 312static inline void tcp_dec_quickack_mode(struct sock *sk,
313 const unsigned int pkts)
1da177e4 314{
463c84b9 315 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 316
463c84b9
ACM		 317 if (icsk->icsk_ack.quick) {
318 if (pkts >= icsk->icsk_ack.quick) {
319 icsk->icsk_ack.quick = 0;
fc6415bc 320 /* Leaving quickack mode we deflate ATO. */
463c84b9 321 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 322 } else
463c84b9 323 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 324 }
325}
326
463c84b9 327extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 328
1da177e4
LT		 329static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
330{
331 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
332}
333
bdf1ee5d
IJ		 334#define TCP_ECN_OK 1
335#define TCP_ECN_QUEUE_CWR 2
336#define TCP_ECN_DEMAND_CWR 4
337
338static __inline__ void
339TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
340{
341 if (sysctl_tcp_ecn && th->ece && th->cwr)
342 inet_rsk(req)->ecn_ok = 1;
343}
344
1da177e4
LT		 345enum tcp_tw_status
346{
347 TCP_TW_SUCCESS = 0,
348 TCP_TW_RST = 1,
349 TCP_TW_ACK = 2,
350 TCP_TW_SYN = 3
351};
352
353
8feaf0c0 354extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 355 struct sk_buff *skb,
8feaf0c0 356 const struct tcphdr *th);
1da177e4
LT		 357
358extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 359 struct request_sock *req,
360 struct request_sock **prev);
1da177e4
LT		 361extern int tcp_child_process(struct sock *parent,
362 struct sock *child,
363 struct sk_buff *skb);
46d0de4e 364extern int tcp_use_frto(struct sock *sk);
1da177e4
LT		 365extern void tcp_enter_frto(struct sock *sk);
366extern void tcp_enter_loss(struct sock *sk, int how);
367extern void tcp_clear_retrans(struct tcp_sock *tp);
368extern void tcp_update_metrics(struct sock *sk);
369
370extern void tcp_close(struct sock *sk,
371 long timeout);
1da177e4
LT		 372extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
373
374extern int tcp_getsockopt(struct sock *sk, int level,
375 int optname,
376 char __user *optval,
377 int __user *optlen);
378extern int tcp_setsockopt(struct sock *sk, int level,
379 int optname, char __user *optval,
380 int optlen);
3fdadf7d
DM		 381extern int compat_tcp_getsockopt(struct sock *sk,
382 int level, int optname,
383 char __user *optval, int __user *optlen);
384extern int compat_tcp_setsockopt(struct sock *sk,
385 int level, int optname,
386 char __user *optval, int optlen);
1da177e4
LT		 387extern void tcp_set_keepalive(struct sock *sk, int val);
388extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
389 struct msghdr *msg,
390 size_t len, int nonblock,
391 int flags, int *addr_len);
392
1da177e4
LT		 393extern void tcp_parse_options(struct sk_buff *skb,
394 struct tcp_options_received *opt_rx,
395 int estab);
396
397/*
398 * TCP v4 functions exported for the inet6 API
399 */
400
8292a17a 401extern void tcp_v4_send_check(struct sock *sk, int len,
1da177e4
LT		 402 struct sk_buff *skb);
403
404extern int tcp_v4_conn_request(struct sock *sk,
405 struct sk_buff *skb);
406
407extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 408 struct request_sock *req,
1da177e4
LT		 409 struct sk_buff *skb);
410
411extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
412 struct sk_buff *skb,
60236fdd 413 struct request_sock *req,
1da177e4
LT		 414 struct dst_entry *dst);
415
416extern int tcp_v4_do_rcv(struct sock *sk,
417 struct sk_buff *skb);
418
419extern int tcp_v4_connect(struct sock *sk,
420 struct sockaddr *uaddr,
421 int addr_len);
422
423extern int tcp_connect(struct sock *sk);
424
425extern struct sk_buff * tcp_make_synack(struct sock *sk,
426 struct dst_entry *dst,
60236fdd 427 struct request_sock *req);
1da177e4
LT		 428
429extern int tcp_disconnect(struct sock *sk, int flags);
430
431extern void tcp_unhash(struct sock *sk);
432
1da177e4
LT		 433/* From syncookies.c */
434extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
435 struct ip_options *opt);
436extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
437 __u16 *mss);
438
439/* tcp_output.c */
440
9e412ba7
IJ		 441extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
442 int nonagle);
443extern int tcp_may_send_now(struct sock *sk);
1da177e4
LT		 444extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
445extern void tcp_xmit_retransmit_queue(struct sock *);
446extern void tcp_simple_retransmit(struct sock *);
447extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 448extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 449
450extern void tcp_send_probe0(struct sock *);
451extern void tcp_send_partial(struct sock *);
452extern int tcp_write_wakeup(struct sock *);
453extern void tcp_send_fin(struct sock *sk);
dd0fc66f 454extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
1da177e4 455extern int tcp_send_synack(struct sock *);
c1b4a7e6 456extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 457extern void tcp_send_ack(struct sock *sk);
458extern void tcp_send_delayed_ack(struct sock *sk);
459
a762a980
DM		 460/* tcp_input.c */
461extern void tcp_cwnd_application_limited(struct sock *sk);
462
1da177e4
LT		 463/* tcp_timer.c */
464extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 465static inline void tcp_clear_xmit_timers(struct sock *sk)
466{
467 inet_csk_clear_xmit_timers(sk);
468}
1da177e4 469
1da177e4
LT		 470extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
471extern unsigned int tcp_current_mss(struct sock *sk, int large);
472
17b085ea 473/* tcp.c */
1da177e4
LT		 474extern void tcp_get_info(struct sock *, struct tcp_info *);
475
476/* Read 'sendfile()'-style from a TCP socket */
477typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
478 unsigned int, size_t);
479extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
480 sk_read_actor_t recv_actor);
481
40efc6fa 482extern void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 483
5d424d5a
JH		 484extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
485extern int tcp_mss_to_mtu(struct sock *sk, int mss);
486extern void tcp_mtup_init(struct sock *sk);
487
40efc6fa 488static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 489{
490 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
491 ntohl(TCP_FLAG_ACK) |
492 snd_wnd);
493}
494
40efc6fa 495static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 496{
497 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
498}
499
9e412ba7 500static inline void tcp_fast_path_check(struct sock *sk)
1da177e4 501{
9e412ba7
IJ		 502 struct tcp_sock *tp = tcp_sk(sk);
503
b03efcfb 504 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 505 tp->rcv_wnd &&
506 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
507 !tp->urg_data)
508 tcp_fast_path_on(tp);
509}
510
511/* Compute the actual receive window we are currently advertising.
512 * Rcv_nxt can be after the window if our peer push more data
513 * than the offered window.
514 */
40efc6fa 515static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 516{
517 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
518
519 if (win < 0)
520 win = 0;
521 return (u32) win;
522}
523
524/* Choose a new window, without checks for shrinking, and without
525 * scaling applied to the result. The caller does these things
526 * if necessary. This is a "raw" window selection.
527 */
528extern u32 __tcp_select_window(struct sock *sk);
529
530/* TCP timestamps are only 32-bits, this causes a slight
531 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 532 * of jiffies in the buffer control blocks below. We decided
533 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 534 * casts with the following macro.
535 */
536#define tcp_time_stamp ((__u32)(jiffies))
537
caa20d9a 538/* This is what the send packet queuing engine uses to pass
1da177e4
LT		 539 * TCP per-packet control information to the transmission
540 * code. We also store the host-order sequence numbers in
541 * here too. This is 36 bytes on 32-bit architectures,
542 * 40 bytes on 64-bit machines, if this grows please adjust
543 * skbuff.h:skbuff->cb[xxx] size appropriately.
544 */
545struct tcp_skb_cb {
546 union {
547 struct inet_skb_parm h4;
548#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
549 struct inet6_skb_parm h6;
550#endif
551 } header; /* For incoming frames */
552 __u32 seq; /* Starting sequence number */
553 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
554 __u32 when; /* used to compute rtt's */
555 __u8 flags; /* TCP header flags. */
556
557 /* NOTE: These must match up to the flags byte in a
558 * real TCP header.
559 */
560#define TCPCB_FLAG_FIN 0x01
561#define TCPCB_FLAG_SYN 0x02
562#define TCPCB_FLAG_RST 0x04
563#define TCPCB_FLAG_PSH 0x08
564#define TCPCB_FLAG_ACK 0x10
565#define TCPCB_FLAG_URG 0x20
566#define TCPCB_FLAG_ECE 0x40
567#define TCPCB_FLAG_CWR 0x80
568
569 __u8 sacked; /* State flags for SACK/FACK. */
570#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
571#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
572#define TCPCB_LOST 0x04 /* SKB is lost */
573#define TCPCB_TAGBITS 0x07 /* All tag bits */
574
575#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
576#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
577
caa20d9a 578#define TCPCB_URG 0x20 /* Urgent pointer advanced here */
1da177e4
LT		 579
580#define TCPCB_AT_TAIL (TCPCB_URG)
581
582 __u16 urg_ptr; /* Valid w/URG flags is set. */
583 __u32 ack_seq; /* Sequence number ACK'd */
584};
585
586#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
587
1da177e4
LT		 588/* Due to TSO, an SKB can be composed of multiple actual
589 * packets. To keep these tracked properly, we use this.
590 */
591static inline int tcp_skb_pcount(const struct sk_buff *skb)
592{
7967168c 593 return skb_shinfo(skb)->gso_segs;
1da177e4
LT		 594}
595
596/* This is valid iff tcp_skb_pcount() > 1. */
597static inline int tcp_skb_mss(const struct sk_buff *skb)
598{
7967168c 599 return skb_shinfo(skb)->gso_size;
1da177e4
LT		 600}
601
af610b4c 602static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
1da177e4
LT		 603{
604 if (*count) {
af610b4c 605 *count -= decr;
1da177e4
LT		 606 if ((int)*count < 0)
607 *count = 0;
608 }
609}
610
af610b4c
IJ		 611static inline void tcp_dec_pcount_approx(__u32 *count,
612 const struct sk_buff *skb)
613{
614 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
615}
616
9e412ba7 617static inline void tcp_packets_out_inc(struct sock *sk,
1da177e4
LT		 618 const struct sk_buff *skb)
619{
9e412ba7 620 struct tcp_sock *tp = tcp_sk(sk);
1da177e4
LT		 621 int orig = tp->packets_out;
622
623 tp->packets_out += tcp_skb_pcount(skb);
624 if (!orig)
3f421baa
ACM		 625 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
626 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 627}
628
629static inline void tcp_packets_out_dec(struct tcp_sock *tp,
630 const struct sk_buff *skb)
631{
632 tp->packets_out -= tcp_skb_pcount(skb);
633}
634
317a76f9
SH		 635/* Events passed to congestion control interface */
636enum tcp_ca_event {
637 CA_EVENT_TX_START, /* first transmit when no packets in flight */
638 CA_EVENT_CWND_RESTART, /* congestion window restart */
639 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
640 CA_EVENT_FRTO, /* fast recovery timeout */
641 CA_EVENT_LOSS, /* loss timeout */
642 CA_EVENT_FAST_ACK, /* in sequence ack */
643 CA_EVENT_SLOW_ACK, /* other ack */
644};
645
646/*
647 * Interface for adding new TCP congestion control handlers
648 */
649#define TCP_CA_NAME_MAX 16
3ff825b2
SH		 650#define TCP_CA_MAX 128
651#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
652
164891aa
SH		 653#define TCP_CONG_NON_RESTRICTED 0x1
654#define TCP_CONG_RTT_STAMP 0x2
655
317a76f9
SH		 656struct tcp_congestion_ops {
657 struct list_head list;
164891aa 658 unsigned long flags;
317a76f9
SH		 659
660 /* initialize private data (optional) */
6687e988 661 void (*init)(struct sock *sk);
317a76f9 662 /* cleanup private data (optional) */
6687e988 663 void (*release)(struct sock *sk);
317a76f9
SH		 664
665 /* return slow start threshold (required) */
6687e988 666 u32 (*ssthresh)(struct sock *sk);
317a76f9 667 /* lower bound for congestion window (optional) */
72dc5b92 668 u32 (*min_cwnd)(const struct sock *sk);
317a76f9 669 /* do new cwnd calculation (required) */
16751347 670 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight, int good_ack);
317a76f9 671 /* call before changing ca_state (optional) */
6687e988 672 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 673 /* call when cwnd event occurs (optional) */
6687e988 674 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 675 /* new value of cwnd after loss (optional) */
6687e988 676 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 677 /* hook for packet ack accounting (optional) */
30cfd0ba 678 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
73c1f4a0 679 /* get info for inet_diag (optional) */
6687e988 680 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 681
682 char name[TCP_CA_NAME_MAX];
683 struct module *owner;
684};
685
686extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
687extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
688
6687e988
ACM		 689extern void tcp_init_congestion_control(struct sock *sk);
690extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 691extern int tcp_set_default_congestion_control(const char *name);
692extern void tcp_get_default_congestion_control(char *name);
3ff825b2 693extern void tcp_get_available_congestion_control(char *buf, size_t len);
ce7bc3bf
SH		 694extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
695extern int tcp_set_allowed_congestion_control(char *allowed);
6687e988 696extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 697extern void tcp_slow_start(struct tcp_sock *tp);
317a76f9 698
5f8ef48d 699extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988 700extern u32 tcp_reno_ssthresh(struct sock *sk);
16751347 701extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight, int flag);
72dc5b92 702extern u32 tcp_reno_min_cwnd(const struct sock *sk);
a8acfbac 703extern struct tcp_congestion_ops tcp_reno;
317a76f9 704
6687e988 705static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 706{
6687e988
ACM		 707 struct inet_connection_sock *icsk = inet_csk(sk);
708
709 if (icsk->icsk_ca_ops->set_state)
710 icsk->icsk_ca_ops->set_state(sk, ca_state);
711 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 712}
713
6687e988 714static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 715{
6687e988
ACM		 716 const struct inet_connection_sock *icsk = inet_csk(sk);
717
718 if (icsk->icsk_ca_ops->cwnd_event)
719 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 720}
721
83ae4088
IJ		 722static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
723{
724 return tp->sacked_out + tp->lost_out;
725}
726
1da177e4
LT		 727/* This determines how many packets are "in the network" to the best
728 * of our knowledge. In many cases it is conservative, but where
729 * detailed information is available from the receiver (via SACK
730 * blocks etc.) we can make more aggressive calculations.
731 *
732 * Use this for decisions involving congestion control, use just
733 * tp->packets_out to determine if the send queue is empty or not.
734 *
735 * Read this equation as:
736 *
737 * "Packets sent once on transmission queue" MINUS
738 * "Packets left network, but not honestly ACKed yet" PLUS
739 * "Packets fast retransmitted"
740 */
40efc6fa 741static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4 742{
83ae4088 743 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1da177e4
LT		 744}
745
1da177e4
LT		 746/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
747 * The exception is rate halving phase, when cwnd is decreasing towards
748 * ssthresh.
749 */
6687e988 750static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 751{
6687e988
ACM		 752 const struct tcp_sock *tp = tcp_sk(sk);
753 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 754 return tp->snd_ssthresh;
755 else
756 return max(tp->snd_ssthresh,
757 ((tp->snd_cwnd >> 1) +
758 (tp->snd_cwnd >> 2)));
759}
760
005903bc 761/* Use define here intentionally to get BUG_ON location shown at the caller */
1b6d427b 762#define tcp_verify_left_out(tp) BUG_ON(tcp_left_out(tp) > tp->packets_out)
1da177e4 763
3cfe3baa 764extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
1da177e4
LT		 765extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
766
767/* Slow start with delack produces 3 packets of burst, so that
768 * it is safe "de facto".
769 */
770static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
771{
772 return 3;
773}
774
f4805ede
SH		 775/* RFC2861 Check whether we are limited by application or congestion window
776 * This is the inverse of cwnd check in tcp_tso_should_defer
777 */
778static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
779{
780 const struct tcp_sock *tp = tcp_sk(sk);
781 u32 left;
782
783 if (in_flight >= tp->snd_cwnd)
784 return 1;
785
bcd76111 786 if (!sk_can_gso(sk))
f4805ede
SH		 787 return 0;
788
789 left = tp->snd_cwnd - in_flight;
790 if (sysctl_tcp_tso_win_divisor)
791 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
792 else
793 return left <= tcp_max_burst(tp);
794}
795
40efc6fa
SH		 796static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
797 const struct sk_buff *skb)
1da177e4
LT		 798{
799 if (skb->len < mss)
800 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
801}
802
9e412ba7 803static inline void tcp_check_probe_timer(struct sock *sk)
1da177e4 804{
9e412ba7 805 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 806 const struct inet_connection_sock *icsk = inet_csk(sk);
9e412ba7 807
463c84b9 808 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 809 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
810 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 811}
812
9e412ba7 813static inline void tcp_push_pending_frames(struct sock *sk)
1da177e4 814{
9e412ba7
IJ		 815 struct tcp_sock *tp = tcp_sk(sk);
816
817 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
1da177e4
LT		 818}
819
40efc6fa 820static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 821{
822 tp->snd_wl1 = seq;
823}
824
40efc6fa 825static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 826{
827 tp->snd_wl1 = seq;
828}
829
1da177e4
LT		 830/*
831 * Calculate(/check) TCP checksum
832 */
ba7808ea
FD		 833static inline __sum16 tcp_v4_check(int len, __be32 saddr,
834 __be32 daddr, __wsum base)
1da177e4
LT		 835{
836 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
837}
838
b51655b9 839static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 840{
fb286bb2 841 return __skb_checksum_complete(skb);
1da177e4
LT		 842}
843
40efc6fa 844static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4 845{
60476372 846 return !skb_csum_unnecessary(skb) &&
1da177e4
LT		 847 __tcp_checksum_complete(skb);
848}
849
850/* Prequeue for VJ style copy to user, combined with checksumming. */
851
40efc6fa 852static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 853{
854 tp->ucopy.task = NULL;
855 tp->ucopy.len = 0;
856 tp->ucopy.memory = 0;
857 skb_queue_head_init(&tp->ucopy.prequeue);
97fc2f08
CL		 858#ifdef CONFIG_NET_DMA
859 tp->ucopy.dma_chan = NULL;
860 tp->ucopy.wakeup = 0;
861 tp->ucopy.pinned_list = NULL;
862 tp->ucopy.dma_cookie = 0;
863#endif
1da177e4
LT		 864}
865
866/* Packet is added to VJ-style prequeue for processing in process
867 * context, if a reader task is waiting. Apparently, this exciting
868 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
869 * failed somewhere. Latency? Burstiness? Well, at least now we will
870 * see, why it failed. 8)8) --ANK
871 *
872 * NOTE: is this not too big to inline?
873 */
40efc6fa 874static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 875{
876 struct tcp_sock *tp = tcp_sk(sk);
877
878 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
879 __skb_queue_tail(&tp->ucopy.prequeue, skb);
880 tp->ucopy.memory += skb->truesize;
881 if (tp->ucopy.memory > sk->sk_rcvbuf) {
882 struct sk_buff *skb1;
883
884 BUG_ON(sock_owned_by_user(sk));
885
886 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
887 sk->sk_backlog_rcv(sk, skb1);
888 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
889 }
890
891 tp->ucopy.memory = 0;
892 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
893 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 894 if (!inet_csk_ack_scheduled(sk))
895 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
3f421baa
ACM		 896 (3 * TCP_RTO_MIN) / 4,
897 TCP_RTO_MAX);
1da177e4
LT		 898 }
899 return 1;
900 }
901 return 0;
902}
903
904
905#undef STATE_TRACE
906
907#ifdef STATE_TRACE
908static const char *statename[]={
909 "Unused","Established","Syn Sent","Syn Recv",
910 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
911 "Close Wait","Last ACK","Listen","Closing"
912};
913#endif
914
40efc6fa 915static inline void tcp_set_state(struct sock *sk, int state)
1da177e4
LT		 916{
917 int oldstate = sk->sk_state;
918
919 switch (state) {
920 case TCP_ESTABLISHED:
921 if (oldstate != TCP_ESTABLISHED)
922 TCP_INC_STATS(TCP_MIB_CURRESTAB);
923 break;
924
925 case TCP_CLOSE:
926 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
927 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
928
929 sk->sk_prot->unhash(sk);
463c84b9 930 if (inet_csk(sk)->icsk_bind_hash &&
1da177e4 931 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
2d8c4ce5 932 inet_put_port(&tcp_hashinfo, sk);
1da177e4
LT		 933 /* fall through */
934 default:
935 if (oldstate==TCP_ESTABLISHED)
936 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
937 }
938
939 /* Change state AFTER socket is unhashed to avoid closed
940 * socket sitting in hash tables.
941 */
942 sk->sk_state = state;
943
944#ifdef STATE_TRACE
945 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
946#endif
947}
948
4ac02bab 949extern void tcp_done(struct sock *sk);
1da177e4 950
40efc6fa 951static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 952{
953 rx_opt->dsack = 0;
954 rx_opt->eff_sacks = 0;
955 rx_opt->num_sacks = 0;
956}
957
1da177e4
LT		 958/* Determine a window scaling and initial window to offer. */
959extern void tcp_select_initial_window(int __space, __u32 mss,
960 __u32 *rcv_wnd, __u32 *window_clamp,
961 int wscale_ok, __u8 *rcv_wscale);
962
963static inline int tcp_win_from_space(int space)
964{
965 return sysctl_tcp_adv_win_scale<=0 ?
966 (space>>(-sysctl_tcp_adv_win_scale)) :
967 space - (space>>sysctl_tcp_adv_win_scale);
968}
969
970/* Note: caller must be prepared to deal with negative returns */
971static inline int tcp_space(const struct sock *sk)
972{
973 return tcp_win_from_space(sk->sk_rcvbuf -
974 atomic_read(&sk->sk_rmem_alloc));
975}
976
977static inline int tcp_full_space(const struct sock *sk)
978{
979 return tcp_win_from_space(sk->sk_rcvbuf);
980}
981
40efc6fa
SH		 982static inline void tcp_openreq_init(struct request_sock *req,
983 struct tcp_options_received *rx_opt,
984 struct sk_buff *skb)
1da177e4 985{
2e6599cb
ACM		 986 struct inet_request_sock *ireq = inet_rsk(req);
987
1da177e4 988 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 989 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 990 req->mss = rx_opt->mss_clamp;
991 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 992 ireq->tstamp_ok = rx_opt->tstamp_ok;
993 ireq->sack_ok = rx_opt->sack_ok;
994 ireq->snd_wscale = rx_opt->snd_wscale;
995 ireq->wscale_ok = rx_opt->wscale_ok;
996 ireq->acked = 0;
997 ireq->ecn_ok = 0;
aa8223c7 998 ireq->rmt_port = tcp_hdr(skb)->source;
1da177e4
LT		 999}
1000
1001extern void tcp_enter_memory_pressure(void);
1002
1da177e4
LT		 1003static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1004{
1005 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1006}
1007
1008static inline int keepalive_time_when(const struct tcp_sock *tp)
1009{
1010 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1011}
1012
463c84b9 1013static inline int tcp_fin_time(const struct sock *sk)
1da177e4 1014{
463c84b9
ACM		 1015 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1016 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 1017
463c84b9
ACM		 1018 if (fin_timeout < (rto << 2) - (rto >> 1))
1019 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 1020
1021 return fin_timeout;
1022}
1023
1024static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1025{
1026 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1027 return 0;
9d729f72 1028 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1da177e4
LT		 1029 return 0;
1030
1031 /* RST segments are not recommended to carry timestamp,
1032 and, if they do, it is recommended to ignore PAWS because
1033 "their cleanup function should take precedence over timestamps."
1034 Certainly, it is mistake. It is necessary to understand the reasons
1035 of this constraint to relax it: if peer reboots, clock may go
1036 out-of-sync and half-open connections will not be reset.
1037 Actually, the problem would be not existing if all
1038 the implementations followed draft about maintaining clock
1039 via reboots. Linux-2.2 DOES NOT!
1040
1041 However, we can relax time bounds for RST segments to MSL.
1042 */
9d729f72 1043 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1da177e4
LT		 1044 return 0;
1045 return 1;
1046}
1047
1da177e4
LT		 1048#define TCP_CHECK_TIMER(sk) do { } while (0)
1049
1da177e4
LT		 1050static inline void tcp_mib_init(void)
1051{
1052 /* See RFC 2012 */
1053 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1054 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1055 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1056 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1057}
1058
6a438bbe
SH		 1059/*from STCP */
1060static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1061 tp->lost_skb_hint = NULL;
1062 tp->scoreboard_skb_hint = NULL;
1063 tp->retransmit_skb_hint = NULL;
1064 tp->forward_skb_hint = NULL;
1065 tp->fastpath_skb_hint = NULL;
1066}
1067
cfb6eeb4
YH		 1068/* MD5 Signature */
1069struct crypto_hash;
1070
1071/* - key database */
1072struct tcp_md5sig_key {
1073 u8 *key;
1074 u8 keylen;
1075};
1076
1077struct tcp4_md5sig_key {
f8ab18d2 1078 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1079 __be32 addr;
1080};
1081
1082struct tcp6_md5sig_key {
f8ab18d2 1083 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1084#if 0
1085 u32 scope_id; /* XXX */
1086#endif
1087 struct in6_addr addr;
1088};
1089
1090/* - sock block */
1091struct tcp_md5sig_info {
1092 struct tcp4_md5sig_key *keys4;
1093#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1094 struct tcp6_md5sig_key *keys6;
1095 u32 entries6;
1096 u32 alloced6;
1097#endif
1098 u32 entries4;
1099 u32 alloced4;
1100};
1101
1102/* - pseudo header */
1103struct tcp4_pseudohdr {
1104 __be32 saddr;
1105 __be32 daddr;
1106 __u8 pad;
1107 __u8 protocol;
1108 __be16 len;
1109};
1110
1111struct tcp6_pseudohdr {
1112 struct in6_addr saddr;
1113 struct in6_addr daddr;
1114 __be32 len;
1115 __be32 protocol; /* including padding */
1116};
1117
1118union tcp_md5sum_block {
1119 struct tcp4_pseudohdr ip4;
1120#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1121 struct tcp6_pseudohdr ip6;
1122#endif
1123};
1124
1125/* - pool: digest algorithm, hash description and scratch buffer */
1126struct tcp_md5sig_pool {
1127 struct hash_desc md5_desc;
1128 union tcp_md5sum_block md5_blk;
1129};
1130
1131#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1132
1133/* - functions */
1134extern int tcp_v4_calc_md5_hash(char *md5_hash,
1135 struct tcp_md5sig_key *key,
1136 struct sock *sk,
1137 struct dst_entry *dst,
1138 struct request_sock *req,
1139 struct tcphdr *th,
1140 int protocol, int tcplen);
1141extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1142 struct sock *addr_sk);
1143
1144extern int tcp_v4_md5_do_add(struct sock *sk,
1145 __be32 addr,
1146 u8 *newkey,
1147 u8 newkeylen);
1148
1149extern int tcp_v4_md5_do_del(struct sock *sk,
8e5200f5 1150 __be32 addr);
cfb6eeb4
YH		 1151
1152extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1153extern void tcp_free_md5sig_pool(void);
1154
1155extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1156extern void __tcp_put_md5sig_pool(void);
1157
1158static inline
1159struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1160{
1161 int cpu = get_cpu();
1162 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1163 if (!ret)
1164 put_cpu();
1165 return ret;
1166}
1167
1168static inline void tcp_put_md5sig_pool(void)
1169{
1170 __tcp_put_md5sig_pool();
1171 put_cpu();
1172}
1173
fe067e8a
DM		 1174/* write queue abstraction */
1175static inline void tcp_write_queue_purge(struct sock *sk)
1176{
1177 struct sk_buff *skb;
1178
1179 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1180 sk_stream_free_skb(sk, skb);
1181 sk_stream_mem_reclaim(sk);
1182}
1183
1184static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1185{
1186 struct sk_buff *skb = sk->sk_write_queue.next;
1187 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1188 return NULL;
1189 return skb;
1190}
1191
1192static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1193{
1194 struct sk_buff *skb = sk->sk_write_queue.prev;
1195 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1196 return NULL;
1197 return skb;
1198}
1199
1200static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1201{
1202 return skb->next;
1203}
1204
1205#define tcp_for_write_queue(skb, sk) \
1206 for (skb = (sk)->sk_write_queue.next; \
1207 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1208 skb = skb->next)
1209
1210#define tcp_for_write_queue_from(skb, sk) \
1211 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1212 skb = skb->next)
1213
1214static inline struct sk_buff *tcp_send_head(struct sock *sk)
1215{
1216 return sk->sk_send_head;
1217}
1218
1219static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1220{
d551e454
IJ		 1221 struct tcp_sock *tp = tcp_sk(sk);
1222
fe067e8a
DM		 1223 sk->sk_send_head = skb->next;
1224 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1225 sk->sk_send_head = NULL;
d551e454
IJ		 1226 /* Don't override Nagle indefinately with F-RTO */
1227 if (tp->frto_counter == 2)
1228 tp->frto_counter = 3;
fe067e8a
DM		 1229}
1230
1231static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1232{
1233 if (sk->sk_send_head == skb_unlinked)
1234 sk->sk_send_head = NULL;
1235}
1236
1237static inline void tcp_init_send_head(struct sock *sk)
1238{
1239 sk->sk_send_head = NULL;
1240}
1241
1242static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1243{
1244 __skb_queue_tail(&sk->sk_write_queue, skb);
1245}
1246
1247static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1248{
1249 __tcp_add_write_queue_tail(sk, skb);
1250
1251 /* Queue it, remembering where we must start sending. */
1252 if (sk->sk_send_head == NULL)
1253 sk->sk_send_head = skb;
1254}
1255
1256static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1257{
1258 __skb_queue_head(&sk->sk_write_queue, skb);
1259}
1260
1261/* Insert buff after skb on the write queue of sk. */
1262static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1263 struct sk_buff *buff,
1264 struct sock *sk)
1265{
1266 __skb_append(skb, buff, &sk->sk_write_queue);
1267}
1268
1269/* Insert skb between prev and next on the write queue of sk. */
1270static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1271 struct sk_buff *skb,
1272 struct sock *sk)
1273{
1274 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1275}
1276
1277static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1278{
1279 __skb_unlink(skb, &sk->sk_write_queue);
1280}
1281
1282static inline int tcp_skb_is_last(const struct sock *sk,
1283 const struct sk_buff *skb)
1284{
1285 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1286}
1287
1288static inline int tcp_write_queue_empty(struct sock *sk)
1289{
1290 return skb_queue_empty(&sk->sk_write_queue);
1291}
1292
1da177e4
LT		 1293/* /proc */
1294enum tcp_seq_states {
1295 TCP_SEQ_STATE_LISTENING,
1296 TCP_SEQ_STATE_OPENREQ,
1297 TCP_SEQ_STATE_ESTABLISHED,
1298 TCP_SEQ_STATE_TIME_WAIT,
1299};
1300
1301struct tcp_seq_afinfo {
1302 struct module *owner;
1303 char *name;
1304 sa_family_t family;
1305 int (*seq_show) (struct seq_file *m, void *v);
1306 struct file_operations *seq_fops;
1307};
1308
1309struct tcp_iter_state {
1310 sa_family_t family;
1311 enum tcp_seq_states state;
1312 struct sock *syn_wait_sk;
1313 int bucket, sbucket, num, uid;
1314 struct seq_operations seq_ops;
1315};
1316
1317extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1318extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1319
20380731
ACM		 1320extern struct request_sock_ops tcp_request_sock_ops;
1321
1322extern int tcp_v4_destroy_sock(struct sock *sk);
1323
a430a43d 1324extern int tcp_v4_gso_send_check(struct sk_buff *skb);
576a30eb 1325extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
f4c50d99 1326
20380731
ACM		 1327#ifdef CONFIG_PROC_FS
1328extern int tcp4_proc_init(void);
1329extern void tcp4_proc_exit(void);
1330#endif
1331
cfb6eeb4
YH		 1332/* TCP af-specific functions */
1333struct tcp_sock_af_ops {
1334#ifdef CONFIG_TCP_MD5SIG
1335 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1336 struct sock *addr_sk);
1337 int (*calc_md5_hash) (char *location,
1338 struct tcp_md5sig_key *md5,
1339 struct sock *sk,
1340 struct dst_entry *dst,
1341 struct request_sock *req,
1342 struct tcphdr *th,
1343 int protocol, int len);
1344 int (*md5_add) (struct sock *sk,
1345 struct sock *addr_sk,
1346 u8 *newkey,
1347 u8 len);
1348 int (*md5_parse) (struct sock *sk,
1349 char __user *optval,
1350 int optlen);
1351#endif
1352};
1353
1354struct tcp_request_sock_ops {
1355#ifdef CONFIG_TCP_MD5SIG
1356 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1357 struct request_sock *req);
1358#endif
1359};
1360
20380731
ACM		 1361extern void tcp_v4_init(struct net_proto_family *ops);
1362extern void tcp_init(void);
1363
1da177e4 1364#endif /* _TCP_H */
Linux 6.x block layer and io_uring tree(s)