?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[TCP]: Restrict congestion control choices.
[linux-2.6/kvm.git] / include / net / tcp.h
blobe1a5d29d0a1fccfa524bec6f4df1e195bbc30c93
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
32 #include <net/inet_connection_sock.h>
33 #include <net/inet_timewait_sock.h>
34 #include <net/inet_hashtables.h>
35 #include <net/checksum.h>
36 #include <net/request_sock.h>
37 #include <net/sock.h>
38 #include <net/snmp.h>
39 #include <net/ip.h>
40 #include <net/tcp_states.h>
41
42 #include <linux/seq_file.h>
43
44 extern struct inet_hashinfo tcp_hashinfo;
45
46 extern atomic_t tcp_orphan_count;
47 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
48
49 #define MAX_TCP_HEADER (128 + MAX_HEADER)
50
51 /*
52 * Never offer a window over 32767 without using window scaling. Some
53 * poor stacks do signed 16bit maths!
54 */
55 #define MAX_TCP_WINDOW 32767U
56
57 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
58 #define TCP_MIN_MSS 88U
59
60 /* Minimal RCV_MSS. */
61 #define TCP_MIN_RCVMSS 536U
62
63 /* The least MTU to use for probing */
64 #define TCP_BASE_MSS 512
65
66 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
67 #define TCP_FASTRETRANS_THRESH 3
68
69 /* Maximal reordering. */
70 #define TCP_MAX_REORDERING 127
71
72 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
73 #define TCP_MAX_QUICKACKS 16U
74
75 /* urg_data states */
76 #define TCP_URG_VALID 0x0100
77 #define TCP_URG_NOTYET 0x0200
78 #define TCP_URG_READ 0x0400
79
80 #define TCP_RETR1 3 /*
81 * This is how many retries it does before it
82 * tries to figure out if the gateway is
83 * down. Minimal RFC value is 3; it corresponds
84 * to ~3sec-8min depending on RTO.
85 */
86
87 #define TCP_RETR2 15 /*
88 * This should take at least
89 * 90 minutes to time out.
90 * RFC1122 says that the limit is 100 sec.
91 * 15 is ~13-30min depending on RTO.
92 */
93
94 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
95 * connection: ~180sec is RFC minimum */
96
97 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
98 * connection: ~180sec is RFC minimum */
99
100
101 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
102 * socket. 7 is ~50sec-16min.
103 */
104
105
106 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
112 * TIME-WAIT timer.
113 */
114
115 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
116 #if HZ >= 100
117 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118 #define TCP_ATO_MIN ((unsigned)(HZ/25))
119 #else
120 #define TCP_DELACK_MIN 4U
121 #define TCP_ATO_MIN 4U
122 #endif
123 #define TCP_RTO_MAX ((unsigned)(120*HZ))
124 #define TCP_RTO_MIN ((unsigned)(HZ/5))
125 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
126
127 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
128 * for local resources.
129 */
130
131 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
132 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
133 #define TCP_KEEPALIVE_INTVL (75*HZ)
134
135 #define MAX_TCP_KEEPIDLE 32767
136 #define MAX_TCP_KEEPINTVL 32767
137 #define MAX_TCP_KEEPCNT 127
138 #define MAX_TCP_SYNCNT 127
139
140 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
141
142 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
143 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
144 * after this time. It should be equal
145 * (or greater than) TCP_TIMEWAIT_LEN
146 * to provide reliability equal to one
147 * provided by timewait state.
148 */
149 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
150 * timestamps. It must be less than
151 * minimal timewait lifetime.
152 */
153 /*
154 * TCP option
155 */
156
157 #define TCPOPT_NOP 1 /* Padding */
158 #define TCPOPT_EOL 0 /* End of options */
159 #define TCPOPT_MSS 2 /* Segment size negotiating */
160 #define TCPOPT_WINDOW 3 /* Window scaling */
161 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
162 #define TCPOPT_SACK 5 /* SACK Block */
163 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
164
165 /*
166 * TCP option lengths
167 */
168
169 #define TCPOLEN_MSS 4
170 #define TCPOLEN_WINDOW 3
171 #define TCPOLEN_SACK_PERM 2
172 #define TCPOLEN_TIMESTAMP 10
173
174 /* But this is what stacks really send out. */
175 #define TCPOLEN_TSTAMP_ALIGNED 12
176 #define TCPOLEN_WSCALE_ALIGNED 4
177 #define TCPOLEN_SACKPERM_ALIGNED 4
178 #define TCPOLEN_SACK_BASE 2
179 #define TCPOLEN_SACK_BASE_ALIGNED 4
180 #define TCPOLEN_SACK_PERBLOCK 8
181
182 /* Flags in tp->nonagle */
183 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
184 #define TCP_NAGLE_CORK 2 /* Socket is corked */
185 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
186
187 extern struct inet_timewait_death_row tcp_death_row;
188
189 /* sysctl variables for tcp */
190 extern int sysctl_tcp_timestamps;
191 extern int sysctl_tcp_window_scaling;
192 extern int sysctl_tcp_sack;
193 extern int sysctl_tcp_fin_timeout;
194 extern int sysctl_tcp_keepalive_time;
195 extern int sysctl_tcp_keepalive_probes;
196 extern int sysctl_tcp_keepalive_intvl;
197 extern int sysctl_tcp_syn_retries;
198 extern int sysctl_tcp_synack_retries;
199 extern int sysctl_tcp_retries1;
200 extern int sysctl_tcp_retries2;
201 extern int sysctl_tcp_orphan_retries;
202 extern int sysctl_tcp_syncookies;
203 extern int sysctl_tcp_retrans_collapse;
204 extern int sysctl_tcp_stdurg;
205 extern int sysctl_tcp_rfc1337;
206 extern int sysctl_tcp_abort_on_overflow;
207 extern int sysctl_tcp_max_orphans;
208 extern int sysctl_tcp_fack;
209 extern int sysctl_tcp_reordering;
210 extern int sysctl_tcp_ecn;
211 extern int sysctl_tcp_dsack;
212 extern int sysctl_tcp_mem[3];
213 extern int sysctl_tcp_wmem[3];
214 extern int sysctl_tcp_rmem[3];
215 extern int sysctl_tcp_app_win;
216 extern int sysctl_tcp_adv_win_scale;
217 extern int sysctl_tcp_tw_reuse;
218 extern int sysctl_tcp_frto;
219 extern int sysctl_tcp_low_latency;
220 extern int sysctl_tcp_dma_copybreak;
221 extern int sysctl_tcp_nometrics_save;
222 extern int sysctl_tcp_moderate_rcvbuf;
223 extern int sysctl_tcp_tso_win_divisor;
224 extern int sysctl_tcp_abc;
225 extern int sysctl_tcp_mtu_probing;
226 extern int sysctl_tcp_base_mss;
227 extern int sysctl_tcp_workaround_signed_windows;
228 extern int sysctl_tcp_slow_start_after_idle;
229
230 extern atomic_t tcp_memory_allocated;
231 extern atomic_t tcp_sockets_allocated;
232 extern int tcp_memory_pressure;
233
234 /*
235 * The next routines deal with comparing 32 bit unsigned ints
236 * and worry about wraparound (automatic with unsigned arithmetic).
237 */
238
239 static inline int before(__u32 seq1, __u32 seq2)
240 {
241 return (__s32)(seq1-seq2) < 0;
242 }
243
244 static inline int after(__u32 seq1, __u32 seq2)
245 {
246 return (__s32)(seq2-seq1) < 0;
247 }
248
249
250 /* is s2<=s1<=s3 ? */
251 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
252 {
253 return seq3 - seq2 >= seq1 - seq2;
254 }
255
256
257 extern struct proto tcp_prot;
258
259 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
260 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
261 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
262 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
263 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
264 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
265 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
266
267 extern void tcp_v4_err(struct sk_buff *skb, u32);
268
269 extern void tcp_shutdown (struct sock *sk, int how);
270
271 extern int tcp_v4_rcv(struct sk_buff *skb);
272
273 extern int tcp_v4_remember_stamp(struct sock *sk);
274
275 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
276
277 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
278 struct msghdr *msg, size_t size);
279 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
280
281 extern int tcp_ioctl(struct sock *sk,
282 int cmd,
283 unsigned long arg);
284
285 extern int tcp_rcv_state_process(struct sock *sk,
286 struct sk_buff *skb,
287 struct tcphdr *th,
288 unsigned len);
289
290 extern int tcp_rcv_established(struct sock *sk,
291 struct sk_buff *skb,
292 struct tcphdr *th,
293 unsigned len);
294
295 extern void tcp_rcv_space_adjust(struct sock *sk);
296
297 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
298
299 extern int tcp_twsk_unique(struct sock *sk,
300 struct sock *sktw, void *twp);
301
302 static inline void tcp_dec_quickack_mode(struct sock *sk,
303 const unsigned int pkts)
304 {
305 struct inet_connection_sock *icsk = inet_csk(sk);
306
307 if (icsk->icsk_ack.quick) {
308 if (pkts >= icsk->icsk_ack.quick) {
309 icsk->icsk_ack.quick = 0;
310 /* Leaving quickack mode we deflate ATO. */
311 icsk->icsk_ack.ato = TCP_ATO_MIN;
312 } else
313 icsk->icsk_ack.quick -= pkts;
314 }
315 }
316
317 extern void tcp_enter_quickack_mode(struct sock *sk);
318
319 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
320 {
321 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
322 }
323
324 enum tcp_tw_status
325 {
326 TCP_TW_SUCCESS = 0,
327 TCP_TW_RST = 1,
328 TCP_TW_ACK = 2,
329 TCP_TW_SYN = 3
330 };
331
332
333 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
334 struct sk_buff *skb,
335 const struct tcphdr *th);
336
337 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
338 struct request_sock *req,
339 struct request_sock **prev);
340 extern int tcp_child_process(struct sock *parent,
341 struct sock *child,
342 struct sk_buff *skb);
343 extern void tcp_enter_frto(struct sock *sk);
344 extern void tcp_enter_loss(struct sock *sk, int how);
345 extern void tcp_clear_retrans(struct tcp_sock *tp);
346 extern void tcp_update_metrics(struct sock *sk);
347
348 extern void tcp_close(struct sock *sk,
349 long timeout);
350 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
351
352 extern int tcp_getsockopt(struct sock *sk, int level,
353 int optname,
354 char __user *optval,
355 int __user *optlen);
356 extern int tcp_setsockopt(struct sock *sk, int level,
357 int optname, char __user *optval,
358 int optlen);
359 extern int compat_tcp_getsockopt(struct sock *sk,
360 int level, int optname,
361 char __user *optval, int __user *optlen);
362 extern int compat_tcp_setsockopt(struct sock *sk,
363 int level, int optname,
364 char __user *optval, int optlen);
365 extern void tcp_set_keepalive(struct sock *sk, int val);
366 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
367 struct msghdr *msg,
368 size_t len, int nonblock,
369 int flags, int *addr_len);
370
371 extern void tcp_parse_options(struct sk_buff *skb,
372 struct tcp_options_received *opt_rx,
373 int estab);
374
375 /*
376 * TCP v4 functions exported for the inet6 API
377 */
378
379 extern void tcp_v4_send_check(struct sock *sk, int len,
380 struct sk_buff *skb);
381
382 extern int tcp_v4_conn_request(struct sock *sk,
383 struct sk_buff *skb);
384
385 extern struct sock * tcp_create_openreq_child(struct sock *sk,
386 struct request_sock *req,
387 struct sk_buff *skb);
388
389 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
390 struct sk_buff *skb,
391 struct request_sock *req,
392 struct dst_entry *dst);
393
394 extern int tcp_v4_do_rcv(struct sock *sk,
395 struct sk_buff *skb);
396
397 extern int tcp_v4_connect(struct sock *sk,
398 struct sockaddr *uaddr,
399 int addr_len);
400
401 extern int tcp_connect(struct sock *sk);
402
403 extern struct sk_buff * tcp_make_synack(struct sock *sk,
404 struct dst_entry *dst,
405 struct request_sock *req);
406
407 extern int tcp_disconnect(struct sock *sk, int flags);
408
409 extern void tcp_unhash(struct sock *sk);
410
411 /* From syncookies.c */
412 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
413 struct ip_options *opt);
414 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
415 __u16 *mss);
416
417 /* tcp_output.c */
418
419 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
420 unsigned int cur_mss, int nonagle);
421 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
422 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
423 extern void tcp_xmit_retransmit_queue(struct sock *);
424 extern void tcp_simple_retransmit(struct sock *);
425 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
426 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
427
428 extern void tcp_send_probe0(struct sock *);
429 extern void tcp_send_partial(struct sock *);
430 extern int tcp_write_wakeup(struct sock *);
431 extern void tcp_send_fin(struct sock *sk);
432 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
433 extern int tcp_send_synack(struct sock *);
434 extern void tcp_push_one(struct sock *, unsigned int mss_now);
435 extern void tcp_send_ack(struct sock *sk);
436 extern void tcp_send_delayed_ack(struct sock *sk);
437
438 /* tcp_input.c */
439 extern void tcp_cwnd_application_limited(struct sock *sk);
440
441 /* tcp_timer.c */
442 extern void tcp_init_xmit_timers(struct sock *);
443 static inline void tcp_clear_xmit_timers(struct sock *sk)
444 {
445 inet_csk_clear_xmit_timers(sk);
446 }
447
448 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
449 extern unsigned int tcp_current_mss(struct sock *sk, int large);
450
451 /* tcp.c */
452 extern void tcp_get_info(struct sock *, struct tcp_info *);
453
454 /* Read 'sendfile()'-style from a TCP socket */
455 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
456 unsigned int, size_t);
457 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
458 sk_read_actor_t recv_actor);
459
460 extern void tcp_initialize_rcv_mss(struct sock *sk);
461
462 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
463 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
464 extern void tcp_mtup_init(struct sock *sk);
465
466 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
467 {
468 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
469 ntohl(TCP_FLAG_ACK) |
470 snd_wnd);
471 }
472
473 static inline void tcp_fast_path_on(struct tcp_sock *tp)
474 {
475 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
476 }
477
478 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
479 {
480 if (skb_queue_empty(&tp->out_of_order_queue) &&
481 tp->rcv_wnd &&
482 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
483 !tp->urg_data)
484 tcp_fast_path_on(tp);
485 }
486
487 /* Compute the actual receive window we are currently advertising.
488 * Rcv_nxt can be after the window if our peer push more data
489 * than the offered window.
490 */
491 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
492 {
493 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
494
495 if (win < 0)
496 win = 0;
497 return (u32) win;
498 }
499
500 /* Choose a new window, without checks for shrinking, and without
501 * scaling applied to the result. The caller does these things
502 * if necessary. This is a "raw" window selection.
503 */
504 extern u32 __tcp_select_window(struct sock *sk);
505
506 /* TCP timestamps are only 32-bits, this causes a slight
507 * complication on 64-bit systems since we store a snapshot
508 * of jiffies in the buffer control blocks below. We decided
509 * to use only the low 32-bits of jiffies and hide the ugly
510 * casts with the following macro.
511 */
512 #define tcp_time_stamp ((__u32)(jiffies))
513
514 /* This is what the send packet queuing engine uses to pass
515 * TCP per-packet control information to the transmission
516 * code. We also store the host-order sequence numbers in
517 * here too. This is 36 bytes on 32-bit architectures,
518 * 40 bytes on 64-bit machines, if this grows please adjust
519 * skbuff.h:skbuff->cb[xxx] size appropriately.
520 */
521 struct tcp_skb_cb {
522 union {
523 struct inet_skb_parm h4;
524 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
525 struct inet6_skb_parm h6;
526 #endif
527 } header; /* For incoming frames */
528 __u32 seq; /* Starting sequence number */
529 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
530 __u32 when; /* used to compute rtt's */
531 __u8 flags; /* TCP header flags. */
532
533 /* NOTE: These must match up to the flags byte in a
534 * real TCP header.
535 */
536 #define TCPCB_FLAG_FIN 0x01
537 #define TCPCB_FLAG_SYN 0x02
538 #define TCPCB_FLAG_RST 0x04
539 #define TCPCB_FLAG_PSH 0x08
540 #define TCPCB_FLAG_ACK 0x10
541 #define TCPCB_FLAG_URG 0x20
542 #define TCPCB_FLAG_ECE 0x40
543 #define TCPCB_FLAG_CWR 0x80
544
545 __u8 sacked; /* State flags for SACK/FACK. */
546 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
547 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
548 #define TCPCB_LOST 0x04 /* SKB is lost */
549 #define TCPCB_TAGBITS 0x07 /* All tag bits */
550
551 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
552 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
553
554 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
555
556 #define TCPCB_AT_TAIL (TCPCB_URG)
557
558 __u16 urg_ptr; /* Valid w/URG flags is set. */
559 __u32 ack_seq; /* Sequence number ACK'd */
560 };
561
562 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
563
564 #include <net/tcp_ecn.h>
565
566 /* Due to TSO, an SKB can be composed of multiple actual
567 * packets. To keep these tracked properly, we use this.
568 */
569 static inline int tcp_skb_pcount(const struct sk_buff *skb)
570 {
571 return skb_shinfo(skb)->gso_segs;
572 }
573
574 /* This is valid iff tcp_skb_pcount() > 1. */
575 static inline int tcp_skb_mss(const struct sk_buff *skb)
576 {
577 return skb_shinfo(skb)->gso_size;
578 }
579
580 static inline void tcp_dec_pcount_approx(__u32 *count,
581 const struct sk_buff *skb)
582 {
583 if (*count) {
584 *count -= tcp_skb_pcount(skb);
585 if ((int)*count < 0)
586 *count = 0;
587 }
588 }
589
590 static inline void tcp_packets_out_inc(struct sock *sk,
591 struct tcp_sock *tp,
592 const struct sk_buff *skb)
593 {
594 int orig = tp->packets_out;
595
596 tp->packets_out += tcp_skb_pcount(skb);
597 if (!orig)
598 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
599 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
600 }
601
602 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
603 const struct sk_buff *skb)
604 {
605 tp->packets_out -= tcp_skb_pcount(skb);
606 }
607
608 /* Events passed to congestion control interface */
609 enum tcp_ca_event {
610 CA_EVENT_TX_START, /* first transmit when no packets in flight */
611 CA_EVENT_CWND_RESTART, /* congestion window restart */
612 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
613 CA_EVENT_FRTO, /* fast recovery timeout */
614 CA_EVENT_LOSS, /* loss timeout */
615 CA_EVENT_FAST_ACK, /* in sequence ack */
616 CA_EVENT_SLOW_ACK, /* other ack */
617 };
618
619 /*
620 * Interface for adding new TCP congestion control handlers
621 */
622 #define TCP_CA_NAME_MAX 16
623 #define TCP_CA_MAX 128
624 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
625
626 struct tcp_congestion_ops {
627 struct list_head list;
628 int non_restricted;
629
630 /* initialize private data (optional) */
631 void (*init)(struct sock *sk);
632 /* cleanup private data (optional) */
633 void (*release)(struct sock *sk);
634
635 /* return slow start threshold (required) */
636 u32 (*ssthresh)(struct sock *sk);
637 /* lower bound for congestion window (optional) */
638 u32 (*min_cwnd)(const struct sock *sk);
639 /* do new cwnd calculation (required) */
640 void (*cong_avoid)(struct sock *sk, u32 ack,
641 u32 rtt, u32 in_flight, int good_ack);
642 /* round trip time sample per acked packet (optional) */
643 void (*rtt_sample)(struct sock *sk, u32 usrtt);
644 /* call before changing ca_state (optional) */
645 void (*set_state)(struct sock *sk, u8 new_state);
646 /* call when cwnd event occurs (optional) */
647 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
648 /* new value of cwnd after loss (optional) */
649 u32 (*undo_cwnd)(struct sock *sk);
650 /* hook for packet ack accounting (optional) */
651 void (*pkts_acked)(struct sock *sk, u32 num_acked);
652 /* get info for inet_diag (optional) */
653 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
654
655 char name[TCP_CA_NAME_MAX];
656 struct module *owner;
657 };
658
659 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
660 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
661
662 extern void tcp_init_congestion_control(struct sock *sk);
663 extern void tcp_cleanup_congestion_control(struct sock *sk);
664 extern int tcp_set_default_congestion_control(const char *name);
665 extern void tcp_get_default_congestion_control(char *name);
666 extern void tcp_get_available_congestion_control(char *buf, size_t len);
667 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
668 extern int tcp_set_allowed_congestion_control(char *allowed);
669 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
670 extern void tcp_slow_start(struct tcp_sock *tp);
671
672 extern struct tcp_congestion_ops tcp_init_congestion_ops;
673 extern u32 tcp_reno_ssthresh(struct sock *sk);
674 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
675 u32 rtt, u32 in_flight, int flag);
676 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
677 extern struct tcp_congestion_ops tcp_reno;
678
679 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
680 {
681 struct inet_connection_sock *icsk = inet_csk(sk);
682
683 if (icsk->icsk_ca_ops->set_state)
684 icsk->icsk_ca_ops->set_state(sk, ca_state);
685 icsk->icsk_ca_state = ca_state;
686 }
687
688 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
689 {
690 const struct inet_connection_sock *icsk = inet_csk(sk);
691
692 if (icsk->icsk_ca_ops->cwnd_event)
693 icsk->icsk_ca_ops->cwnd_event(sk, event);
694 }
695
696 /* This determines how many packets are "in the network" to the best
697 * of our knowledge. In many cases it is conservative, but where
698 * detailed information is available from the receiver (via SACK
699 * blocks etc.) we can make more aggressive calculations.
700 *
701 * Use this for decisions involving congestion control, use just
702 * tp->packets_out to determine if the send queue is empty or not.
703 *
704 * Read this equation as:
705 *
706 * "Packets sent once on transmission queue" MINUS
707 * "Packets left network, but not honestly ACKed yet" PLUS
708 * "Packets fast retransmitted"
709 */
710 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
711 {
712 return (tp->packets_out - tp->left_out + tp->retrans_out);
713 }
714
715 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
716 * The exception is rate halving phase, when cwnd is decreasing towards
717 * ssthresh.
718 */
719 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
720 {
721 const struct tcp_sock *tp = tcp_sk(sk);
722 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
723 return tp->snd_ssthresh;
724 else
725 return max(tp->snd_ssthresh,
726 ((tp->snd_cwnd >> 1) +
727 (tp->snd_cwnd >> 2)));
728 }
729
730 static inline void tcp_sync_left_out(struct tcp_sock *tp)
731 {
732 if (tp->rx_opt.sack_ok &&
733 (tp->sacked_out >= tp->packets_out - tp->lost_out))
734 tp->sacked_out = tp->packets_out - tp->lost_out;
735 tp->left_out = tp->sacked_out + tp->lost_out;
736 }
737
738 extern void tcp_enter_cwr(struct sock *sk);
739 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
740
741 /* Slow start with delack produces 3 packets of burst, so that
742 * it is safe "de facto".
743 */
744 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
745 {
746 return 3;
747 }
748
749 /* RFC2861 Check whether we are limited by application or congestion window
750 * This is the inverse of cwnd check in tcp_tso_should_defer
751 */
752 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
753 {
754 const struct tcp_sock *tp = tcp_sk(sk);
755 u32 left;
756
757 if (in_flight >= tp->snd_cwnd)
758 return 1;
759
760 if (!sk_can_gso(sk))
761 return 0;
762
763 left = tp->snd_cwnd - in_flight;
764 if (sysctl_tcp_tso_win_divisor)
765 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
766 else
767 return left <= tcp_max_burst(tp);
768 }
769
770 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
771 const struct sk_buff *skb)
772 {
773 if (skb->len < mss)
774 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
775 }
776
777 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
778 {
779 const struct inet_connection_sock *icsk = inet_csk(sk);
780 if (!tp->packets_out && !icsk->icsk_pending)
781 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
782 icsk->icsk_rto, TCP_RTO_MAX);
783 }
784
785 static inline void tcp_push_pending_frames(struct sock *sk,
786 struct tcp_sock *tp)
787 {
788 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
789 }
790
791 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
792 {
793 tp->snd_wl1 = seq;
794 }
795
796 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
797 {
798 tp->snd_wl1 = seq;
799 }
800
801 /*
802 * Calculate(/check) TCP checksum
803 */
804 static inline u16 tcp_v4_check(struct tcphdr *th, int len,
805 unsigned long saddr, unsigned long daddr,
806 unsigned long base)
807 {
808 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
809 }
810
811 static inline int __tcp_checksum_complete(struct sk_buff *skb)
812 {
813 return __skb_checksum_complete(skb);
814 }
815
816 static inline int tcp_checksum_complete(struct sk_buff *skb)
817 {
818 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
819 __tcp_checksum_complete(skb);
820 }
821
822 /* Prequeue for VJ style copy to user, combined with checksumming. */
823
824 static inline void tcp_prequeue_init(struct tcp_sock *tp)
825 {
826 tp->ucopy.task = NULL;
827 tp->ucopy.len = 0;
828 tp->ucopy.memory = 0;
829 skb_queue_head_init(&tp->ucopy.prequeue);
830 #ifdef CONFIG_NET_DMA
831 tp->ucopy.dma_chan = NULL;
832 tp->ucopy.wakeup = 0;
833 tp->ucopy.pinned_list = NULL;
834 tp->ucopy.dma_cookie = 0;
835 #endif
836 }
837
838 /* Packet is added to VJ-style prequeue for processing in process
839 * context, if a reader task is waiting. Apparently, this exciting
840 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
841 * failed somewhere. Latency? Burstiness? Well, at least now we will
842 * see, why it failed. 8)8) --ANK
843 *
844 * NOTE: is this not too big to inline?
845 */
846 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
847 {
848 struct tcp_sock *tp = tcp_sk(sk);
849
850 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
851 __skb_queue_tail(&tp->ucopy.prequeue, skb);
852 tp->ucopy.memory += skb->truesize;
853 if (tp->ucopy.memory > sk->sk_rcvbuf) {
854 struct sk_buff *skb1;
855
856 BUG_ON(sock_owned_by_user(sk));
857
858 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
859 sk->sk_backlog_rcv(sk, skb1);
860 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
861 }
862
863 tp->ucopy.memory = 0;
864 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
865 wake_up_interruptible(sk->sk_sleep);
866 if (!inet_csk_ack_scheduled(sk))
867 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
868 (3 * TCP_RTO_MIN) / 4,
869 TCP_RTO_MAX);
870 }
871 return 1;
872 }
873 return 0;
874 }
875
876
877 #undef STATE_TRACE
878
879 #ifdef STATE_TRACE
880 static const char *statename[]={
881 "Unused","Established","Syn Sent","Syn Recv",
882 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
883 "Close Wait","Last ACK","Listen","Closing"
884 };
885 #endif
886
887 static inline void tcp_set_state(struct sock *sk, int state)
888 {
889 int oldstate = sk->sk_state;
890
891 switch (state) {
892 case TCP_ESTABLISHED:
893 if (oldstate != TCP_ESTABLISHED)
894 TCP_INC_STATS(TCP_MIB_CURRESTAB);
895 break;
896
897 case TCP_CLOSE:
898 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
899 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
900
901 sk->sk_prot->unhash(sk);
902 if (inet_csk(sk)->icsk_bind_hash &&
903 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
904 inet_put_port(&tcp_hashinfo, sk);
905 /* fall through */
906 default:
907 if (oldstate==TCP_ESTABLISHED)
908 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
909 }
910
911 /* Change state AFTER socket is unhashed to avoid closed
912 * socket sitting in hash tables.
913 */
914 sk->sk_state = state;
915
916 #ifdef STATE_TRACE
917 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
918 #endif
919 }
920
921 static inline void tcp_done(struct sock *sk)
922 {
923 if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
924 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
925
926 tcp_set_state(sk, TCP_CLOSE);
927 tcp_clear_xmit_timers(sk);
928
929 sk->sk_shutdown = SHUTDOWN_MASK;
930
931 if (!sock_flag(sk, SOCK_DEAD))
932 sk->sk_state_change(sk);
933 else
934 inet_csk_destroy_sock(sk);
935 }
936
937 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
938 {
939 rx_opt->dsack = 0;
940 rx_opt->eff_sacks = 0;
941 rx_opt->num_sacks = 0;
942 }
943
944 /* Determine a window scaling and initial window to offer. */
945 extern void tcp_select_initial_window(int __space, __u32 mss,
946 __u32 *rcv_wnd, __u32 *window_clamp,
947 int wscale_ok, __u8 *rcv_wscale);
948
949 static inline int tcp_win_from_space(int space)
950 {
951 return sysctl_tcp_adv_win_scale<=0 ?
952 (space>>(-sysctl_tcp_adv_win_scale)) :
953 space - (space>>sysctl_tcp_adv_win_scale);
954 }
955
956 /* Note: caller must be prepared to deal with negative returns */
957 static inline int tcp_space(const struct sock *sk)
958 {
959 return tcp_win_from_space(sk->sk_rcvbuf -
960 atomic_read(&sk->sk_rmem_alloc));
961 }
962
963 static inline int tcp_full_space(const struct sock *sk)
964 {
965 return tcp_win_from_space(sk->sk_rcvbuf);
966 }
967
968 static inline void tcp_openreq_init(struct request_sock *req,
969 struct tcp_options_received *rx_opt,
970 struct sk_buff *skb)
971 {
972 struct inet_request_sock *ireq = inet_rsk(req);
973
974 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
975 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
976 req->mss = rx_opt->mss_clamp;
977 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
978 ireq->tstamp_ok = rx_opt->tstamp_ok;
979 ireq->sack_ok = rx_opt->sack_ok;
980 ireq->snd_wscale = rx_opt->snd_wscale;
981 ireq->wscale_ok = rx_opt->wscale_ok;
982 ireq->acked = 0;
983 ireq->ecn_ok = 0;
984 ireq->rmt_port = skb->h.th->source;
985 }
986
987 extern void tcp_enter_memory_pressure(void);
988
989 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
990 {
991 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
992 }
993
994 static inline int keepalive_time_when(const struct tcp_sock *tp)
995 {
996 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
997 }
998
999 static inline int tcp_fin_time(const struct sock *sk)
1000 {
1001 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1002 const int rto = inet_csk(sk)->icsk_rto;
1003
1004 if (fin_timeout < (rto << 2) - (rto >> 1))
1005 fin_timeout = (rto << 2) - (rto >> 1);
1006
1007 return fin_timeout;
1008 }
1009
1010 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1011 {
1012 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1013 return 0;
1014 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1015 return 0;
1016
1017 /* RST segments are not recommended to carry timestamp,
1018 and, if they do, it is recommended to ignore PAWS because
1019 "their cleanup function should take precedence over timestamps."
1020 Certainly, it is mistake. It is necessary to understand the reasons
1021 of this constraint to relax it: if peer reboots, clock may go
1022 out-of-sync and half-open connections will not be reset.
1023 Actually, the problem would be not existing if all
1024 the implementations followed draft about maintaining clock
1025 via reboots. Linux-2.2 DOES NOT!
1026
1027 However, we can relax time bounds for RST segments to MSL.
1028 */
1029 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1030 return 0;
1031 return 1;
1032 }
1033
1034 #define TCP_CHECK_TIMER(sk) do { } while (0)
1035
1036 static inline int tcp_use_frto(const struct sock *sk)
1037 {
1038 const struct tcp_sock *tp = tcp_sk(sk);
1039
1040 /* F-RTO must be activated in sysctl and there must be some
1041 * unsent new data, and the advertised window should allow
1042 * sending it.
1043 */
1044 return (sysctl_tcp_frto && sk->sk_send_head &&
1045 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1046 tp->snd_una + tp->snd_wnd));
1047 }
1048
1049 static inline void tcp_mib_init(void)
1050 {
1051 /* See RFC 2012 */
1052 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1053 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1054 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1055 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1056 }
1057
1058 /*from STCP */
1059 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1060 tp->lost_skb_hint = NULL;
1061 tp->scoreboard_skb_hint = NULL;
1062 tp->retransmit_skb_hint = NULL;
1063 tp->forward_skb_hint = NULL;
1064 tp->fastpath_skb_hint = NULL;
1065 }
1066
1067 /* /proc */
1068 enum tcp_seq_states {
1069 TCP_SEQ_STATE_LISTENING,
1070 TCP_SEQ_STATE_OPENREQ,
1071 TCP_SEQ_STATE_ESTABLISHED,
1072 TCP_SEQ_STATE_TIME_WAIT,
1073 };
1074
1075 struct tcp_seq_afinfo {
1076 struct module *owner;
1077 char *name;
1078 sa_family_t family;
1079 int (*seq_show) (struct seq_file *m, void *v);
1080 struct file_operations *seq_fops;
1081 };
1082
1083 struct tcp_iter_state {
1084 sa_family_t family;
1085 enum tcp_seq_states state;
1086 struct sock *syn_wait_sk;
1087 int bucket, sbucket, num, uid;
1088 struct seq_operations seq_ops;
1089 };
1090
1091 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1092 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1093
1094 extern struct request_sock_ops tcp_request_sock_ops;
1095
1096 extern int tcp_v4_destroy_sock(struct sock *sk);
1097
1098 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1099 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1100
1101 #ifdef CONFIG_PROC_FS
1102 extern int tcp4_proc_init(void);
1103 extern void tcp4_proc_exit(void);
1104 #endif
1105
1106 extern void tcp_v4_init(struct net_proto_family *ops);
1107 extern void tcp_init(void);
1108
1109 #endif /* _TCP_H */
kernel-based virtual machine