tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 *
10 * Authors: Ross Biro
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
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>
29#include <linux/skbuff.h>
30#include <linux/dmaengine.h>
31#include <linux/crypto.h>
32#include <linux/cryptohash.h>
33
34#include <net/inet_connection_sock.h>
35#include <net/inet_timewait_sock.h>
36#include <net/inet_hashtables.h>
37#include <net/checksum.h>
38#include <net/request_sock.h>
39#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
42#include <net/tcp_states.h>
43#include <net/inet_ecn.h>
44
45#include <linux/seq_file.h>
46
47extern struct inet_hashinfo tcp_hashinfo;
48
49extern struct percpu_counter tcp_orphan_count;
50extern void tcp_time_wait(struct sock *sk, int state, int timeo);
51
52#define MAX_TCP_HEADER (128 + MAX_HEADER)
53#define MAX_TCP_OPTION_SPACE 40
54
55/*
56 * Never offer a window over 32767 without using window scaling. Some
57 * poor stacks do signed 16bit maths!
58 */
59#define MAX_TCP_WINDOW 32767U
60
61/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
62#define TCP_MIN_MSS 88U
63
64/* Minimal RCV_MSS. */
65#define TCP_MIN_RCVMSS 536U
66
67/* The least MTU to use for probing */
68#define TCP_BASE_MSS 512
69
70/* After receiving this amount of duplicate ACKs fast retransmit starts. */
71#define TCP_FASTRETRANS_THRESH 3
72
73/* Maximal reordering. */
74#define TCP_MAX_REORDERING 127
75
76/* Maximal number of ACKs sent quickly to accelerate slow-start. */
77#define TCP_MAX_QUICKACKS 16U
78
79/* urg_data states */
80#define TCP_URG_VALID 0x0100
81#define TCP_URG_NOTYET 0x0200
82#define TCP_URG_READ 0x0400
83
84#define TCP_RETR1 3 /*
85 * This is how many retries it does before it
86 * tries to figure out if the gateway is
87 * down. Minimal RFC value is 3; it corresponds
88 * to ~3sec-8min depending on RTO.
89 */
90
91#define TCP_RETR2 15 /*
92 * This should take at least
93 * 90 minutes to time out.
94 * RFC1122 says that the limit is 100 sec.
95 * 15 is ~13-30min depending on RTO.
96 */
97
98#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
99 * connection: ~180sec is RFC minimum */
100
101#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
102 * connection: ~180sec is RFC minimum */
103
104
105#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
106 * socket. 7 is ~50sec-16min.
107 */
108
109
110#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
111 * state, about 60 seconds */
112#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
113 /* BSD style FIN_WAIT2 deadlock breaker.
114 * It used to be 3min, new value is 60sec,
115 * to combine FIN-WAIT-2 timeout with
116 * TIME-WAIT timer.
117 */
118
119#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
120#if HZ >= 100
121#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
122#define TCP_ATO_MIN ((unsigned)(HZ/25))
123#else
124#define TCP_DELACK_MIN 4U
125#define TCP_ATO_MIN 4U
126#endif
127#define TCP_RTO_MAX ((unsigned)(120*HZ))
128#define TCP_RTO_MIN ((unsigned)(HZ/5))
129#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
130
131#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
132 * for local resources.
133 */
134
135#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
136#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
137#define TCP_KEEPALIVE_INTVL (75*HZ)
138
139#define MAX_TCP_KEEPIDLE 32767
140#define MAX_TCP_KEEPINTVL 32767
141#define MAX_TCP_KEEPCNT 127
142#define MAX_TCP_SYNCNT 127
143
144#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
145
146#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
147#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
148 * after this time. It should be equal
149 * (or greater than) TCP_TIMEWAIT_LEN
150 * to provide reliability equal to one
151 * provided by timewait state.
152 */
153#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
154 * timestamps. It must be less than
155 * minimal timewait lifetime.
156 */
157/*
158 * TCP option
159 */
160
161#define TCPOPT_NOP 1 /* Padding */
162#define TCPOPT_EOL 0 /* End of options */
163#define TCPOPT_MSS 2 /* Segment size negotiating */
164#define TCPOPT_WINDOW 3 /* Window scaling */
165#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
166#define TCPOPT_SACK 5 /* SACK Block */
167#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
168#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
169
170/*
171 * TCP option lengths
172 */
173
174#define TCPOLEN_MSS 4
175#define TCPOLEN_WINDOW 3
176#define TCPOLEN_SACK_PERM 2
177#define TCPOLEN_TIMESTAMP 10
178#define TCPOLEN_MD5SIG 18
179
180/* But this is what stacks really send out. */
181#define TCPOLEN_TSTAMP_ALIGNED 12
182#define TCPOLEN_WSCALE_ALIGNED 4
183#define TCPOLEN_SACKPERM_ALIGNED 4
184#define TCPOLEN_SACK_BASE 2
185#define TCPOLEN_SACK_BASE_ALIGNED 4
186#define TCPOLEN_SACK_PERBLOCK 8
187#define TCPOLEN_MD5SIG_ALIGNED 20
188#define TCPOLEN_MSS_ALIGNED 4
189
190/* Flags in tp->nonagle */
191#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
192#define TCP_NAGLE_CORK 2 /* Socket is corked */
193#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
194
195extern struct inet_timewait_death_row tcp_death_row;
196
197/* sysctl variables for tcp */
198extern int sysctl_tcp_timestamps;
199extern int sysctl_tcp_window_scaling;
200extern int sysctl_tcp_sack;
201extern int sysctl_tcp_fin_timeout;
202extern int sysctl_tcp_keepalive_time;
203extern int sysctl_tcp_keepalive_probes;
204extern int sysctl_tcp_keepalive_intvl;
205extern int sysctl_tcp_syn_retries;
206extern int sysctl_tcp_synack_retries;
207extern int sysctl_tcp_retries1;
208extern int sysctl_tcp_retries2;
209extern int sysctl_tcp_orphan_retries;
210extern int sysctl_tcp_syncookies;
211extern int sysctl_tcp_retrans_collapse;
212extern int sysctl_tcp_stdurg;
213extern int sysctl_tcp_rfc1337;
214extern int sysctl_tcp_abort_on_overflow;
215extern int sysctl_tcp_max_orphans;
216extern int sysctl_tcp_fack;
217extern int sysctl_tcp_reordering;
218extern int sysctl_tcp_ecn;
219extern int sysctl_tcp_dsack;
220extern int sysctl_tcp_mem[3];
221extern int sysctl_tcp_wmem[3];
222extern int sysctl_tcp_rmem[3];
223extern int sysctl_tcp_app_win;
224extern int sysctl_tcp_adv_win_scale;
225extern int sysctl_tcp_tw_reuse;
226extern int sysctl_tcp_frto;
227extern int sysctl_tcp_frto_response;
228extern int sysctl_tcp_low_latency;
229extern int sysctl_tcp_dma_copybreak;
230extern int sysctl_tcp_nometrics_save;
231extern int sysctl_tcp_moderate_rcvbuf;
232extern int sysctl_tcp_tso_win_divisor;
233extern int sysctl_tcp_abc;
234extern int sysctl_tcp_mtu_probing;
235extern int sysctl_tcp_base_mss;
236extern int sysctl_tcp_workaround_signed_windows;
237extern int sysctl_tcp_slow_start_after_idle;
238extern int sysctl_tcp_max_ssthresh;
239
240extern atomic_t tcp_memory_allocated;
241extern struct percpu_counter tcp_sockets_allocated;
242extern int tcp_memory_pressure;
243
244/*
245 * The next routines deal with comparing 32 bit unsigned ints
246 * and worry about wraparound (automatic with unsigned arithmetic).
247 */
248
249static inline int before(__u32 seq1, __u32 seq2)
250{
251 return (__s32)(seq1-seq2) < 0;
252}
253#define after(seq2, seq1) before(seq1, seq2)
254
255/* is s2<=s1<=s3 ? */
256static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
257{
258 return seq3 - seq2 >= seq1 - seq2;
259}
260
261static inline int tcp_too_many_orphans(struct sock *sk, int num)
262{
263 return (num > sysctl_tcp_max_orphans) ||
264 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
265 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
266}
267
268extern struct proto tcp_prot;
269
270#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
271#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
272#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
273#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
274
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
283extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
284
285extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
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
305extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
306
307extern int tcp_twsk_unique(struct sock *sk,
308 struct sock *sktw, void *twp);
309
310extern void tcp_twsk_destructor(struct sock *sk);
311
312extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
313 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
314
315static inline void tcp_dec_quickack_mode(struct sock *sk,
316 const unsigned int pkts)
317{
318 struct inet_connection_sock *icsk = inet_csk(sk);
319
320 if (icsk->icsk_ack.quick) {
321 if (pkts >= icsk->icsk_ack.quick) {
322 icsk->icsk_ack.quick = 0;
323 /* Leaving quickack mode we deflate ATO. */
324 icsk->icsk_ack.ato = TCP_ATO_MIN;
325 } else
326 icsk->icsk_ack.quick -= pkts;
327 }
328}
329
330extern void tcp_enter_quickack_mode(struct sock *sk);
331
332static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
333{
334 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
335}
336
337#define TCP_ECN_OK 1
338#define TCP_ECN_QUEUE_CWR 2
339#define TCP_ECN_DEMAND_CWR 4
340
341static __inline__ void
342TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
343{
344 if (sysctl_tcp_ecn && th->ece && th->cwr)
345 inet_rsk(req)->ecn_ok = 1;
346}
347
348enum tcp_tw_status
349{
350 TCP_TW_SUCCESS = 0,
351 TCP_TW_RST = 1,
352 TCP_TW_ACK = 2,
353 TCP_TW_SYN = 3
354};
355
356
357extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
358 struct sk_buff *skb,
359 const struct tcphdr *th);
360
361extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
362 struct request_sock *req,
363 struct request_sock **prev);
364extern int tcp_child_process(struct sock *parent,
365 struct sock *child,
366 struct sk_buff *skb);
367extern int tcp_use_frto(struct sock *sk);
368extern void tcp_enter_frto(struct sock *sk);
369extern void tcp_enter_loss(struct sock *sk, int how);
370extern void tcp_clear_retrans(struct tcp_sock *tp);
371extern void tcp_update_metrics(struct sock *sk);
372
373extern void tcp_close(struct sock *sk,
374 long timeout);
375extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
376
377extern int tcp_getsockopt(struct sock *sk, int level,
378 int optname,
379 char __user *optval,
380 int __user *optlen);
381extern int tcp_setsockopt(struct sock *sk, int level,
382 int optname, char __user *optval,
383 int optlen);
384extern int compat_tcp_getsockopt(struct sock *sk,
385 int level, int optname,
386 char __user *optval, int __user *optlen);
387extern int compat_tcp_setsockopt(struct sock *sk,
388 int level, int optname,
389 char __user *optval, int optlen);
390extern void tcp_set_keepalive(struct sock *sk, int val);
391extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
392 struct msghdr *msg,
393 size_t len, int nonblock,
394 int flags, int *addr_len);
395
396extern void tcp_parse_options(struct sk_buff *skb,
397 struct tcp_options_received *opt_rx,
398 int estab);
399
400extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
401
402/*
403 * TCP v4 functions exported for the inet6 API
404 */
405
406extern void tcp_v4_send_check(struct sock *sk, int len,
407 struct sk_buff *skb);
408
409extern int tcp_v4_conn_request(struct sock *sk,
410 struct sk_buff *skb);
411
412extern struct sock * tcp_create_openreq_child(struct sock *sk,
413 struct request_sock *req,
414 struct sk_buff *skb);
415
416extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
417 struct sk_buff *skb,
418 struct request_sock *req,
419 struct dst_entry *dst);
420
421extern int tcp_v4_do_rcv(struct sock *sk,
422 struct sk_buff *skb);
423
424extern int tcp_v4_connect(struct sock *sk,
425 struct sockaddr *uaddr,
426 int addr_len);
427
428extern int tcp_connect(struct sock *sk);
429
430extern struct sk_buff * tcp_make_synack(struct sock *sk,
431 struct dst_entry *dst,
432 struct request_sock *req);
433
434extern int tcp_disconnect(struct sock *sk, int flags);
435
436
437/* From syncookies.c */
438extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
439extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
440 struct ip_options *opt);
441extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
442 __u16 *mss);
443
444extern __u32 cookie_init_timestamp(struct request_sock *req);
445extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
446
447/* From net/ipv6/syncookies.c */
448extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
449extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
450 __u16 *mss);
451
452/* tcp_output.c */
453
454extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
455 int nonagle);
456extern int tcp_may_send_now(struct sock *sk);
457extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
458extern void tcp_xmit_retransmit_queue(struct sock *);
459extern void tcp_simple_retransmit(struct sock *);
460extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
461extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
462
463extern void tcp_send_probe0(struct sock *);
464extern void tcp_send_partial(struct sock *);
465extern int tcp_write_wakeup(struct sock *);
466extern void tcp_send_fin(struct sock *sk);
467extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
468extern int tcp_send_synack(struct sock *);
469extern void tcp_push_one(struct sock *, unsigned int mss_now);
470extern void tcp_send_ack(struct sock *sk);
471extern void tcp_send_delayed_ack(struct sock *sk);
472
473/* tcp_input.c */
474extern void tcp_cwnd_application_limited(struct sock *sk);
475
476/* tcp_timer.c */
477extern void tcp_init_xmit_timers(struct sock *);
478static inline void tcp_clear_xmit_timers(struct sock *sk)
479{
480 inet_csk_clear_xmit_timers(sk);
481}
482
483extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
484extern unsigned int tcp_current_mss(struct sock *sk, int large);
485
486/* tcp.c */
487extern void tcp_get_info(struct sock *, struct tcp_info *);
488
489/* Read 'sendfile()'-style from a TCP socket */
490typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
491 unsigned int, size_t);
492extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
493 sk_read_actor_t recv_actor);
494
495extern void tcp_initialize_rcv_mss(struct sock *sk);
496
497extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
498extern int tcp_mss_to_mtu(struct sock *sk, int mss);
499extern void tcp_mtup_init(struct sock *sk);
500
501static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
502{
503 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
504 ntohl(TCP_FLAG_ACK) |
505 snd_wnd);
506}
507
508static inline void tcp_fast_path_on(struct tcp_sock *tp)
509{
510 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
511}
512
513static inline void tcp_fast_path_check(struct sock *sk)
514{
515 struct tcp_sock *tp = tcp_sk(sk);
516
517 if (skb_queue_empty(&tp->out_of_order_queue) &&
518 tp->rcv_wnd &&
519 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
520 !tp->urg_data)
521 tcp_fast_path_on(tp);
522}
523
524/* Compute the actual receive window we are currently advertising.
525 * Rcv_nxt can be after the window if our peer push more data
526 * than the offered window.
527 */
528static inline u32 tcp_receive_window(const struct tcp_sock *tp)
529{
530 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
531
532 if (win < 0)
533 win = 0;
534 return (u32) win;
535}
536
537/* Choose a new window, without checks for shrinking, and without
538 * scaling applied to the result. The caller does these things
539 * if necessary. This is a "raw" window selection.
540 */
541extern u32 __tcp_select_window(struct sock *sk);
542
543/* TCP timestamps are only 32-bits, this causes a slight
544 * complication on 64-bit systems since we store a snapshot
545 * of jiffies in the buffer control blocks below. We decided
546 * to use only the low 32-bits of jiffies and hide the ugly
547 * casts with the following macro.
548 */
549#define tcp_time_stamp ((__u32)(jiffies))
550
551/* This is what the send packet queuing engine uses to pass
552 * TCP per-packet control information to the transmission
553 * code. We also store the host-order sequence numbers in
554 * here too. This is 36 bytes on 32-bit architectures,
555 * 40 bytes on 64-bit machines, if this grows please adjust
556 * skbuff.h:skbuff->cb[xxx] size appropriately.
557 */
558struct tcp_skb_cb {
559 union {
560 struct inet_skb_parm h4;
561#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
562 struct inet6_skb_parm h6;
563#endif
564 } header; /* For incoming frames */
565 __u32 seq; /* Starting sequence number */
566 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
567 __u32 when; /* used to compute rtt's */
568 __u8 flags; /* TCP header flags. */
569
570 /* NOTE: These must match up to the flags byte in a
571 * real TCP header.
572 */
573#define TCPCB_FLAG_FIN 0x01
574#define TCPCB_FLAG_SYN 0x02
575#define TCPCB_FLAG_RST 0x04
576#define TCPCB_FLAG_PSH 0x08
577#define TCPCB_FLAG_ACK 0x10
578#define TCPCB_FLAG_URG 0x20
579#define TCPCB_FLAG_ECE 0x40
580#define TCPCB_FLAG_CWR 0x80
581
582 __u8 sacked; /* State flags for SACK/FACK. */
583#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
584#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
585#define TCPCB_LOST 0x04 /* SKB is lost */
586#define TCPCB_TAGBITS 0x07 /* All tag bits */
587
588#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
589#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
590
591 __u32 ack_seq; /* Sequence number ACK'd */
592};
593
594#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
595
596/* Due to TSO, an SKB can be composed of multiple actual
597 * packets. To keep these tracked properly, we use this.
598 */
599static inline int tcp_skb_pcount(const struct sk_buff *skb)
600{
601 return skb_shinfo(skb)->gso_segs;
602}
603
604/* This is valid iff tcp_skb_pcount() > 1. */
605static inline int tcp_skb_mss(const struct sk_buff *skb)
606{
607 return skb_shinfo(skb)->gso_size;
608}
609
610static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
611{
612 if (*count) {
613 *count -= decr;
614 if ((int)*count < 0)
615 *count = 0;
616 }
617}
618
619static inline void tcp_dec_pcount_approx(__u32 *count,
620 const struct sk_buff *skb)
621{
622 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
623}
624
625/* Events passed to congestion control interface */
626enum tcp_ca_event {
627 CA_EVENT_TX_START, /* first transmit when no packets in flight */
628 CA_EVENT_CWND_RESTART, /* congestion window restart */
629 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
630 CA_EVENT_FRTO, /* fast recovery timeout */
631 CA_EVENT_LOSS, /* loss timeout */
632 CA_EVENT_FAST_ACK, /* in sequence ack */
633 CA_EVENT_SLOW_ACK, /* other ack */
634};
635
636/*
637 * Interface for adding new TCP congestion control handlers
638 */
639#define TCP_CA_NAME_MAX 16
640#define TCP_CA_MAX 128
641#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
642
643#define TCP_CONG_NON_RESTRICTED 0x1
644#define TCP_CONG_RTT_STAMP 0x2
645
646struct tcp_congestion_ops {
647 struct list_head list;
648 unsigned long flags;
649
650 /* initialize private data (optional) */
651 void (*init)(struct sock *sk);
652 /* cleanup private data (optional) */
653 void (*release)(struct sock *sk);
654
655 /* return slow start threshold (required) */
656 u32 (*ssthresh)(struct sock *sk);
657 /* lower bound for congestion window (optional) */
658 u32 (*min_cwnd)(const struct sock *sk);
659 /* do new cwnd calculation (required) */
660 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
661 /* call before changing ca_state (optional) */
662 void (*set_state)(struct sock *sk, u8 new_state);
663 /* call when cwnd event occurs (optional) */
664 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
665 /* new value of cwnd after loss (optional) */
666 u32 (*undo_cwnd)(struct sock *sk);
667 /* hook for packet ack accounting (optional) */
668 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
669 /* get info for inet_diag (optional) */
670 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
671
672 char name[TCP_CA_NAME_MAX];
673 struct module *owner;
674};
675
676extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
677extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
678
679extern void tcp_init_congestion_control(struct sock *sk);
680extern void tcp_cleanup_congestion_control(struct sock *sk);
681extern int tcp_set_default_congestion_control(const char *name);
682extern void tcp_get_default_congestion_control(char *name);
683extern void tcp_get_available_congestion_control(char *buf, size_t len);
684extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
685extern int tcp_set_allowed_congestion_control(char *allowed);
686extern int tcp_set_congestion_control(struct sock *sk, const char *name);
687extern void tcp_slow_start(struct tcp_sock *tp);
688
689extern struct tcp_congestion_ops tcp_init_congestion_ops;
690extern u32 tcp_reno_ssthresh(struct sock *sk);
691extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
692extern u32 tcp_reno_min_cwnd(const struct sock *sk);
693extern struct tcp_congestion_ops tcp_reno;
694
695static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
696{
697 struct inet_connection_sock *icsk = inet_csk(sk);
698
699 if (icsk->icsk_ca_ops->set_state)
700 icsk->icsk_ca_ops->set_state(sk, ca_state);
701 icsk->icsk_ca_state = ca_state;
702}
703
704static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
705{
706 const struct inet_connection_sock *icsk = inet_csk(sk);
707
708 if (icsk->icsk_ca_ops->cwnd_event)
709 icsk->icsk_ca_ops->cwnd_event(sk, event);
710}
711
712/* These functions determine how the current flow behaves in respect of SACK
713 * handling. SACK is negotiated with the peer, and therefore it can vary
714 * between different flows.
715 *
716 * tcp_is_sack - SACK enabled
717 * tcp_is_reno - No SACK
718 * tcp_is_fack - FACK enabled, implies SACK enabled
719 */
720static inline int tcp_is_sack(const struct tcp_sock *tp)
721{
722 return tp->rx_opt.sack_ok;
723}
724
725static inline int tcp_is_reno(const struct tcp_sock *tp)
726{
727 return !tcp_is_sack(tp);
728}
729
730static inline int tcp_is_fack(const struct tcp_sock *tp)
731{
732 return tp->rx_opt.sack_ok & 2;
733}
734
735static inline void tcp_enable_fack(struct tcp_sock *tp)
736{
737 tp->rx_opt.sack_ok |= 2;
738}
739
740static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
741{
742 return tp->sacked_out + tp->lost_out;
743}
744
745/* This determines how many packets are "in the network" to the best
746 * of our knowledge. In many cases it is conservative, but where
747 * detailed information is available from the receiver (via SACK
748 * blocks etc.) we can make more aggressive calculations.
749 *
750 * Use this for decisions involving congestion control, use just
751 * tp->packets_out to determine if the send queue is empty or not.
752 *
753 * Read this equation as:
754 *
755 * "Packets sent once on transmission queue" MINUS
756 * "Packets left network, but not honestly ACKed yet" PLUS
757 * "Packets fast retransmitted"
758 */
759static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
760{
761 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
762}
763
764/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
765 * The exception is rate halving phase, when cwnd is decreasing towards
766 * ssthresh.
767 */
768static inline __u32 tcp_current_ssthresh(const struct sock *sk)
769{
770 const struct tcp_sock *tp = tcp_sk(sk);
771 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
772 return tp->snd_ssthresh;
773 else
774 return max(tp->snd_ssthresh,
775 ((tp->snd_cwnd >> 1) +
776 (tp->snd_cwnd >> 2)));
777}
778
779/* Use define here intentionally to get WARN_ON location shown at the caller */
780#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
781
782extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
783extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
784
785/* Slow start with delack produces 3 packets of burst, so that
786 * it is safe "de facto". This will be the default - same as
787 * the default reordering threshold - but if reordering increases,
788 * we must be able to allow cwnd to burst at least this much in order
789 * to not pull it back when holes are filled.
790 */
791static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
792{
793 return tp->reordering;
794}
795
796/* Returns end sequence number of the receiver's advertised window */
797static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
798{
799 return tp->snd_una + tp->snd_wnd;
800}
801extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
802
803static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
804 const struct sk_buff *skb)
805{
806 if (skb->len < mss)
807 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
808}
809
810static inline void tcp_check_probe_timer(struct sock *sk)
811{
812 struct tcp_sock *tp = tcp_sk(sk);
813 const struct inet_connection_sock *icsk = inet_csk(sk);
814
815 if (!tp->packets_out && !icsk->icsk_pending)
816 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
817 icsk->icsk_rto, TCP_RTO_MAX);
818}
819
820static inline void tcp_push_pending_frames(struct sock *sk)
821{
822 struct tcp_sock *tp = tcp_sk(sk);
823
824 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
825}
826
827static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
828{
829 tp->snd_wl1 = seq;
830}
831
832static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
833{
834 tp->snd_wl1 = seq;
835}
836
837/*
838 * Calculate(/check) TCP checksum
839 */
840static inline __sum16 tcp_v4_check(int len, __be32 saddr,
841 __be32 daddr, __wsum base)
842{
843 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
844}
845
846static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
847{
848 return __skb_checksum_complete(skb);
849}
850
851static inline int tcp_checksum_complete(struct sk_buff *skb)
852{
853 return !skb_csum_unnecessary(skb) &&
854 __tcp_checksum_complete(skb);
855}
856
857/* Prequeue for VJ style copy to user, combined with checksumming. */
858
859static inline void tcp_prequeue_init(struct tcp_sock *tp)
860{
861 tp->ucopy.task = NULL;
862 tp->ucopy.len = 0;
863 tp->ucopy.memory = 0;
864 skb_queue_head_init(&tp->ucopy.prequeue);
865#ifdef CONFIG_NET_DMA
866 tp->ucopy.dma_chan = NULL;
867 tp->ucopy.wakeup = 0;
868 tp->ucopy.pinned_list = NULL;
869 tp->ucopy.dma_cookie = 0;
870#endif
871}
872
873/* Packet is added to VJ-style prequeue for processing in process
874 * context, if a reader task is waiting. Apparently, this exciting
875 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
876 * failed somewhere. Latency? Burstiness? Well, at least now we will
877 * see, why it failed. 8)8) --ANK
878 *
879 * NOTE: is this not too big to inline?
880 */
881static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
882{
883 struct tcp_sock *tp = tcp_sk(sk);
884
885 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
886 __skb_queue_tail(&tp->ucopy.prequeue, skb);
887 tp->ucopy.memory += skb->truesize;
888 if (tp->ucopy.memory > sk->sk_rcvbuf) {
889 struct sk_buff *skb1;
890
891 BUG_ON(sock_owned_by_user(sk));
892
893 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
894 sk_backlog_rcv(sk, skb1);
895 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED);
896 }
897
898 tp->ucopy.memory = 0;
899 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
900 wake_up_interruptible(sk->sk_sleep);
901 if (!inet_csk_ack_scheduled(sk))
902 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
903 (3 * TCP_RTO_MIN) / 4,
904 TCP_RTO_MAX);
905 }
906 return 1;
907 }
908 return 0;
909}
910
911
912#undef STATE_TRACE
913
914#ifdef STATE_TRACE
915static const char *statename[]={
916 "Unused","Established","Syn Sent","Syn Recv",
917 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
918 "Close Wait","Last ACK","Listen","Closing"
919};
920#endif
921extern void tcp_set_state(struct sock *sk, int state);
922
923extern void tcp_done(struct sock *sk);
924
925static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
926{
927 rx_opt->dsack = 0;
928 rx_opt->eff_sacks = 0;
929 rx_opt->num_sacks = 0;
930}
931
932/* Determine a window scaling and initial window to offer. */
933extern void tcp_select_initial_window(int __space, __u32 mss,
934 __u32 *rcv_wnd, __u32 *window_clamp,
935 int wscale_ok, __u8 *rcv_wscale);
936
937static inline int tcp_win_from_space(int space)
938{
939 return sysctl_tcp_adv_win_scale<=0 ?
940 (space>>(-sysctl_tcp_adv_win_scale)) :
941 space - (space>>sysctl_tcp_adv_win_scale);
942}
943
944/* Note: caller must be prepared to deal with negative returns */
945static inline int tcp_space(const struct sock *sk)
946{
947 return tcp_win_from_space(sk->sk_rcvbuf -
948 atomic_read(&sk->sk_rmem_alloc));
949}
950
951static inline int tcp_full_space(const struct sock *sk)
952{
953 return tcp_win_from_space(sk->sk_rcvbuf);
954}
955
956static inline void tcp_openreq_init(struct request_sock *req,
957 struct tcp_options_received *rx_opt,
958 struct sk_buff *skb)
959{
960 struct inet_request_sock *ireq = inet_rsk(req);
961
962 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
963 req->cookie_ts = 0;
964 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
965 req->mss = rx_opt->mss_clamp;
966 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
967 ireq->tstamp_ok = rx_opt->tstamp_ok;
968 ireq->sack_ok = rx_opt->sack_ok;
969 ireq->snd_wscale = rx_opt->snd_wscale;
970 ireq->wscale_ok = rx_opt->wscale_ok;
971 ireq->acked = 0;
972 ireq->ecn_ok = 0;
973 ireq->rmt_port = tcp_hdr(skb)->source;
974 ireq->loc_port = tcp_hdr(skb)->dest;
975}
976
977extern void tcp_enter_memory_pressure(struct sock *sk);
978
979static inline int keepalive_intvl_when(const struct tcp_sock *tp)
980{
981 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
982}
983
984static inline int keepalive_time_when(const struct tcp_sock *tp)
985{
986 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
987}
988
989static inline int tcp_fin_time(const struct sock *sk)
990{
991 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
992 const int rto = inet_csk(sk)->icsk_rto;
993
994 if (fin_timeout < (rto << 2) - (rto >> 1))
995 fin_timeout = (rto << 2) - (rto >> 1);
996
997 return fin_timeout;
998}
999
1000static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1001{
1002 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1003 return 0;
1004 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1005 return 0;
1006
1007 /* RST segments are not recommended to carry timestamp,
1008 and, if they do, it is recommended to ignore PAWS because
1009 "their cleanup function should take precedence over timestamps."
1010 Certainly, it is mistake. It is necessary to understand the reasons
1011 of this constraint to relax it: if peer reboots, clock may go
1012 out-of-sync and half-open connections will not be reset.
1013 Actually, the problem would be not existing if all
1014 the implementations followed draft about maintaining clock
1015 via reboots. Linux-2.2 DOES NOT!
1016
1017 However, we can relax time bounds for RST segments to MSL.
1018 */
1019 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1020 return 0;
1021 return 1;
1022}
1023
1024#define TCP_CHECK_TIMER(sk) do { } while (0)
1025
1026static inline void tcp_mib_init(struct net *net)
1027{
1028 /* See RFC 2012 */
1029 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1030 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1031 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1032 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1033}
1034
1035/* from STCP */
1036static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1037{
1038 tp->lost_skb_hint = NULL;
1039 tp->scoreboard_skb_hint = NULL;
1040}
1041
1042static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1043{
1044 tcp_clear_retrans_hints_partial(tp);
1045 tp->retransmit_skb_hint = NULL;
1046}
1047
1048/* MD5 Signature */
1049struct crypto_hash;
1050
1051/* - key database */
1052struct tcp_md5sig_key {
1053 u8 *key;
1054 u8 keylen;
1055};
1056
1057struct tcp4_md5sig_key {
1058 struct tcp_md5sig_key base;
1059 __be32 addr;
1060};
1061
1062struct tcp6_md5sig_key {
1063 struct tcp_md5sig_key base;
1064#if 0
1065 u32 scope_id; /* XXX */
1066#endif
1067 struct in6_addr addr;
1068};
1069
1070/* - sock block */
1071struct tcp_md5sig_info {
1072 struct tcp4_md5sig_key *keys4;
1073#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1074 struct tcp6_md5sig_key *keys6;
1075 u32 entries6;
1076 u32 alloced6;
1077#endif
1078 u32 entries4;
1079 u32 alloced4;
1080};
1081
1082/* - pseudo header */
1083struct tcp4_pseudohdr {
1084 __be32 saddr;
1085 __be32 daddr;
1086 __u8 pad;
1087 __u8 protocol;
1088 __be16 len;
1089};
1090
1091struct tcp6_pseudohdr {
1092 struct in6_addr saddr;
1093 struct in6_addr daddr;
1094 __be32 len;
1095 __be32 protocol; /* including padding */
1096};
1097
1098union tcp_md5sum_block {
1099 struct tcp4_pseudohdr ip4;
1100#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1101 struct tcp6_pseudohdr ip6;
1102#endif
1103};
1104
1105/* - pool: digest algorithm, hash description and scratch buffer */
1106struct tcp_md5sig_pool {
1107 struct hash_desc md5_desc;
1108 union tcp_md5sum_block md5_blk;
1109};
1110
1111#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1112
1113/* - functions */
1114extern int tcp_v4_md5_hash_skb(char *md5_hash,
1115 struct tcp_md5sig_key *key,
1116 struct sock *sk,
1117 struct request_sock *req,
1118 struct sk_buff *skb);
1119
1120extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1121 struct sock *addr_sk);
1122
1123extern int tcp_v4_md5_do_add(struct sock *sk,
1124 __be32 addr,
1125 u8 *newkey,
1126 u8 newkeylen);
1127
1128extern int tcp_v4_md5_do_del(struct sock *sk,
1129 __be32 addr);
1130
1131#ifdef CONFIG_TCP_MD5SIG
1132#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1133 &(struct tcp_md5sig_key) { \
1134 .key = (twsk)->tw_md5_key, \
1135 .keylen = (twsk)->tw_md5_keylen, \
1136 } : NULL)
1137#else
1138#define tcp_twsk_md5_key(twsk) NULL
1139#endif
1140
1141extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1142extern void tcp_free_md5sig_pool(void);
1143
1144extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1145extern void __tcp_put_md5sig_pool(void);
1146extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1147extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1148 unsigned header_len);
1149extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1150 struct tcp_md5sig_key *key);
1151
1152static inline
1153struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1154{
1155 int cpu = get_cpu();
1156 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1157 if (!ret)
1158 put_cpu();
1159 return ret;
1160}
1161
1162static inline void tcp_put_md5sig_pool(void)
1163{
1164 __tcp_put_md5sig_pool();
1165 put_cpu();
1166}
1167
1168/* write queue abstraction */
1169static inline void tcp_write_queue_purge(struct sock *sk)
1170{
1171 struct sk_buff *skb;
1172
1173 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1174 sk_wmem_free_skb(sk, skb);
1175 sk_mem_reclaim(sk);
1176}
1177
1178static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1179{
1180 return skb_peek(&sk->sk_write_queue);
1181}
1182
1183static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1184{
1185 return skb_peek_tail(&sk->sk_write_queue);
1186}
1187
1188static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1189{
1190 return skb_queue_next(&sk->sk_write_queue, skb);
1191}
1192
1193static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1194{
1195 return skb_queue_prev(&sk->sk_write_queue, skb);
1196}
1197
1198#define tcp_for_write_queue(skb, sk) \
1199 skb_queue_walk(&(sk)->sk_write_queue, skb)
1200
1201#define tcp_for_write_queue_from(skb, sk) \
1202 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1203
1204#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1205 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1206
1207static inline struct sk_buff *tcp_send_head(struct sock *sk)
1208{
1209 return sk->sk_send_head;
1210}
1211
1212static inline bool tcp_skb_is_last(const struct sock *sk,
1213 const struct sk_buff *skb)
1214{
1215 return skb_queue_is_last(&sk->sk_write_queue, skb);
1216}
1217
1218static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1219{
1220 if (tcp_skb_is_last(sk, skb))
1221 sk->sk_send_head = NULL;
1222 else
1223 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1224}
1225
1226static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1227{
1228 if (sk->sk_send_head == skb_unlinked)
1229 sk->sk_send_head = NULL;
1230}
1231
1232static inline void tcp_init_send_head(struct sock *sk)
1233{
1234 sk->sk_send_head = NULL;
1235}
1236
1237static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1238{
1239 __skb_queue_tail(&sk->sk_write_queue, skb);
1240}
1241
1242static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1243{
1244 __tcp_add_write_queue_tail(sk, skb);
1245
1246 /* Queue it, remembering where we must start sending. */
1247 if (sk->sk_send_head == NULL) {
1248 sk->sk_send_head = skb;
1249
1250 if (tcp_sk(sk)->highest_sack == NULL)
1251 tcp_sk(sk)->highest_sack = skb;
1252 }
1253}
1254
1255static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1256{
1257 __skb_queue_head(&sk->sk_write_queue, skb);
1258}
1259
1260/* Insert buff after skb on the write queue of sk. */
1261static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1262 struct sk_buff *buff,
1263 struct sock *sk)
1264{
1265 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1266}
1267
1268/* Insert new before skb on the write queue of sk. */
1269static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1270 struct sk_buff *skb,
1271 struct sock *sk)
1272{
1273 __skb_queue_before(&sk->sk_write_queue, skb, new);
1274
1275 if (sk->sk_send_head == skb)
1276 sk->sk_send_head = new;
1277}
1278
1279static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1280{
1281 __skb_unlink(skb, &sk->sk_write_queue);
1282}
1283
1284static inline int tcp_write_queue_empty(struct sock *sk)
1285{
1286 return skb_queue_empty(&sk->sk_write_queue);
1287}
1288
1289/* Start sequence of the highest skb with SACKed bit, valid only if
1290 * sacked > 0 or when the caller has ensured validity by itself.
1291 */
1292static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1293{
1294 if (!tp->sacked_out)
1295 return tp->snd_una;
1296
1297 if (tp->highest_sack == NULL)
1298 return tp->snd_nxt;
1299
1300 return TCP_SKB_CB(tp->highest_sack)->seq;
1301}
1302
1303static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1304{
1305 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1306 tcp_write_queue_next(sk, skb);
1307}
1308
1309static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1310{
1311 return tcp_sk(sk)->highest_sack;
1312}
1313
1314static inline void tcp_highest_sack_reset(struct sock *sk)
1315{
1316 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1317}
1318
1319/* Called when old skb is about to be deleted (to be combined with new skb) */
1320static inline void tcp_highest_sack_combine(struct sock *sk,
1321 struct sk_buff *old,
1322 struct sk_buff *new)
1323{
1324 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1325 tcp_sk(sk)->highest_sack = new;
1326}
1327
1328/* /proc */
1329enum tcp_seq_states {
1330 TCP_SEQ_STATE_LISTENING,
1331 TCP_SEQ_STATE_OPENREQ,
1332 TCP_SEQ_STATE_ESTABLISHED,
1333 TCP_SEQ_STATE_TIME_WAIT,
1334};
1335
1336struct tcp_seq_afinfo {
1337 char *name;
1338 sa_family_t family;
1339 struct file_operations seq_fops;
1340 struct seq_operations seq_ops;
1341};
1342
1343struct tcp_iter_state {
1344 struct seq_net_private p;
1345 sa_family_t family;
1346 enum tcp_seq_states state;
1347 struct sock *syn_wait_sk;
1348 int bucket, sbucket, num, uid;
1349};
1350
1351extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1352extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1353
1354extern struct request_sock_ops tcp_request_sock_ops;
1355extern struct request_sock_ops tcp6_request_sock_ops;
1356
1357extern void tcp_v4_destroy_sock(struct sock *sk);
1358
1359extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1360extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1361extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1362 struct sk_buff *skb);
1363extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1364 struct sk_buff *skb);
1365extern int tcp_gro_complete(struct sk_buff *skb);
1366extern int tcp4_gro_complete(struct sk_buff *skb);
1367
1368#ifdef CONFIG_PROC_FS
1369extern int tcp4_proc_init(void);
1370extern void tcp4_proc_exit(void);
1371#endif
1372
1373/* TCP af-specific functions */
1374struct tcp_sock_af_ops {
1375#ifdef CONFIG_TCP_MD5SIG
1376 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1377 struct sock *addr_sk);
1378 int (*calc_md5_hash) (char *location,
1379 struct tcp_md5sig_key *md5,
1380 struct sock *sk,
1381 struct request_sock *req,
1382 struct sk_buff *skb);
1383 int (*md5_add) (struct sock *sk,
1384 struct sock *addr_sk,
1385 u8 *newkey,
1386 u8 len);
1387 int (*md5_parse) (struct sock *sk,
1388 char __user *optval,
1389 int optlen);
1390#endif
1391};
1392
1393struct tcp_request_sock_ops {
1394#ifdef CONFIG_TCP_MD5SIG
1395 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1396 struct request_sock *req);
1397#endif
1398};
1399
1400extern void tcp_v4_init(void);
1401extern void tcp_init(void);
1402
1403#endif /* _TCP_H */