tcp: Reimplement TCP_FASTKEEP socket option using per-pcb keepidle
[dragonfly.git] / sys / netinet / tcp_var.h
CommitLineData
984263bc 1/*
66d6c637
JH
2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
f23061d4 4 *
66d6c637
JH
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
f23061d4 7 *
66d6c637
JH
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
f23061d4 19 *
66d6c637
JH
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34/*
984263bc
MD
35 * Copyright (c) 1982, 1986, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)tcp_var.h 8.4 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_var.h,v 1.56.2.13 2003/02/03 02:34:07 hsu Exp $
e3873585 68 * $DragonFly: src/sys/netinet/tcp_var.h,v 1.42 2008/10/27 02:56:30 sephe Exp $
984263bc
MD
69 */
70
71#ifndef _NETINET_TCP_VAR_H_
72#define _NETINET_TCP_VAR_H_
73
1bd40720 74#ifndef _NETINET_IN_PCB_H_
984263bc 75#include <netinet/in_pcb.h> /* needed for in_conninfo, inp_gen_t */
1bd40720
MD
76#endif
77#ifndef _NETINET_TCP_H_
78#include <netinet/tcp.h>
79#endif
984263bc
MD
80
81/*
82 * Kernel variables for tcp.
83 */
8acdb67c 84extern int tcp_do_rfc1323;
48a4676d 85extern int tcp_low_rtobase;
8acdb67c
JH
86extern int tcp_do_sack;
87extern int tcp_do_smartsack;
88extern int tcp_aggregate_acks;
984263bc
MD
89
90/* TCP segment queue entry */
91struct tseg_qent {
92 LIST_ENTRY(tseg_qent) tqe_q;
93 int tqe_len; /* TCP segment data length */
94 struct tcphdr *tqe_th; /* a pointer to tcp header */
95 struct mbuf *tqe_m; /* mbuf contains packet */
96};
97LIST_HEAD(tsegqe_head, tseg_qent);
3edf7c37
RG
98extern int tcp_reass_maxseg;
99extern int tcp_reass_qsize;
984263bc
MD
100#ifdef MALLOC_DECLARE
101MALLOC_DECLARE(M_TSEGQ);
102#endif
103
104struct tcptemp {
105 u_char tt_ipgen[40]; /* the size must be of max ip header, now IPv6 */
106 struct tcphdr tt_t;
107};
108
109#define tcp6cb tcpcb /* for KAME src sync over BSD*'s */
110
91489f6b
JH
111struct raw_sackblock { /* covers [start, end) */
112 tcp_seq rblk_start;
113 tcp_seq rblk_end;
114};
115
116/* maximum number of SACK blocks that will fit in the TCP option space */
117#define MAX_SACK_REPORT_BLOCKS 4
118
119TAILQ_HEAD(sackblock_list, sackblock);
120
121struct scoreboard {
122 int nblocks;
123 struct sackblock_list sackblocks;
124 tcp_seq lostseq; /* passed SACK lost test */
125 struct sackblock *lastfound; /* search hint */
006af0db 126 struct sackblock *freecache; /* one slot free block cache */
91489f6b
JH
127};
128
0f758523
SZ
129struct netmsg_tcp_timer;
130
984263bc
MD
131/*
132 * Tcp control block, one per tcp; fields:
133 * Organized for 16 byte cacheline efficiency.
134 */
135struct tcpcb {
136 struct tsegqe_head t_segq;
137 int t_dupacks; /* consecutive dup acks recd */
2b1ce38a 138 int tt_cpu; /* sanity check the cpu */
c1d3b892 139
a48c5dd5
SZ
140 struct tcp_callout *tt_rexmt; /* retransmit timer */
141 struct tcp_callout *tt_persist;/* retransmit persistence */
142 struct tcp_callout *tt_keep; /* keepalive */
143 struct tcp_callout *tt_2msl; /* 2*msl TIME_WAIT timer */
144 struct tcp_callout *tt_delack; /* delayed ACK timer */
0f758523 145 struct netmsg_tcp_timer *tt_msg; /* timer message */
984263bc
MD
146
147 struct inpcb *t_inpcb; /* back pointer to internet pcb */
148 int t_state; /* state of this connection */
149 u_int t_flags;
8819433a
JH
150#define TF_ACKNOW 0x00000001 /* ack peer immediately */
151#define TF_DELACK 0x00000002 /* ack, but try to delay it */
152#define TF_NODELAY 0x00000004 /* don't delay packets to coalesce */
153#define TF_NOOPT 0x00000008 /* don't use tcp options */
154#define TF_SENTFIN 0x00000010 /* have sent FIN */
155#define TF_REQ_SCALE 0x00000020 /* have/will request window scaling */
156#define TF_RCVD_SCALE 0x00000040 /* other side has requested scaling */
157#define TF_REQ_TSTMP 0x00000080 /* have/will request timestamps */
158#define TF_RCVD_TSTMP 0x00000100 /* a timestamp was received in SYN */
91489f6b 159#define TF_SACK_PERMITTED 0x00000200 /* other side said I could SACK */
8819433a
JH
160#define TF_NEEDSYN 0x00000400 /* send SYN (implicit state) */
161#define TF_NEEDFIN 0x00000800 /* send FIN (implicit state) */
162#define TF_NOPUSH 0x00001000 /* don't push */
f7b29de5 163#define TF_LISTEN 0x00002000 /* listen(2) has been called */
2a0c7c52 164#define TF_SIGNATURE 0x00004000 /* require MD5 digests (RFC2385) */
fb8d5c6d 165#define TF_UNUSED01 0x00008000 /* unused */
8819433a 166#define TF_MORETOCOME 0x00010000 /* More data to be appended to sock */
d5082e3d 167#define TF_UNUSED00 0x00020000 /* unused */
8819433a
JH
168#define TF_LASTIDLE 0x00040000 /* connection was previously idle */
169#define TF_RXWIN0SENT 0x00080000 /* sent a receiver win 0 in response */
170#define TF_FASTRECOVERY 0x00100000 /* in NewReno Fast Recovery */
171#define TF_WASFRECOVERY 0x00200000 /* was in NewReno Fast Recovery */
172#define TF_FIRSTACCACK 0x00400000 /* Look for 1st acceptable ACK. */
173#define TF_FASTREXMT 0x00800000 /* Did Fast Retransmit. */
174#define TF_EARLYREXMT 0x01000000 /* Did Early (Fast) Retransmit. */
c1d3b892 175#define TF_FORCE 0x02000000 /* Set if forcing out a byte */
2b1ce38a 176#define TF_ONOUTPUTQ 0x04000000 /* on t_outputq list */
91489f6b
JH
177#define TF_DUPSEG 0x08000000 /* last seg a duplicate */
178#define TF_ENCLOSESEG 0x10000000 /* enclosing SACK block */
179#define TF_SACKLEFT 0x20000000 /* send SACK blocks from left side */
efca2b8e 180#define TF_KEEPALIVE 0x40000000 /* temporary keepalive */
46e92930 181#define TF_RXRESIZED 0x80000000 /* rcvbuf was resized */
c1d3b892 182 tcp_seq snd_up; /* send urgent pointer */
3d127502 183 u_long snd_last; /* time last data were sent */
984263bc
MD
184
185 tcp_seq snd_una; /* send unacknowledged */
c1d3b892 186 tcp_seq snd_recover; /* for use with NewReno Fast Recovery */
984263bc 187 tcp_seq snd_max; /* highest sequence number sent;
8acdb67c 188 * used to recognize retransmits */
984263bc 189 tcp_seq snd_nxt; /* send next */
984263bc
MD
190
191 tcp_seq snd_wl1; /* window update seg seq number */
192 tcp_seq snd_wl2; /* window update seg ack number */
193 tcp_seq iss; /* initial send sequence number */
194 tcp_seq irs; /* initial receive sequence number */
195
196 tcp_seq rcv_nxt; /* receive next */
197 tcp_seq rcv_adv; /* advertised window */
198 u_long rcv_wnd; /* receive window */
199 tcp_seq rcv_up; /* receive urgent pointer */
200
201 u_long snd_wnd; /* send window */
202 u_long snd_cwnd; /* congestion-controlled window */
8acdb67c 203 u_long snd_wacked; /* bytes acked in one send window */
984263bc
MD
204 u_long snd_ssthresh; /* snd_cwnd size threshold for
205 * for slow start exponential to
8acdb67c
JH
206 * linear switch */
207
208 int t_rxtcur; /* current retransmit value (ticks) */
209 u_int t_maxseg; /* maximum segment size */
210 int t_srtt; /* smoothed round-trip time */
211 int t_rttvar; /* variance in round-trip time */
984263bc
MD
212
213 u_int t_maxopd; /* mss plus options */
214
3d127502 215 u_long t_rcvtime; /* reception inactivity time */
984263bc
MD
216 u_long t_starttime; /* time connection was established */
217 int t_rtttime; /* round trip time */
218 tcp_seq t_rtseq; /* sequence number being timed */
219
984263bc
MD
220 int t_rxtshift; /* log(2) of rexmt exp. backoff */
221 u_int t_rttmin; /* minimum rtt allowed */
222 u_int t_rttbest; /* best rtt we've seen */
223 u_long t_rttupdated; /* number of times rtt sampled */
224 u_long max_sndwnd; /* largest window peer has offered */
225
226 int t_softerror; /* possible error not yet reported */
227/* out-of-band data */
228 char t_oobflags; /* have some */
229 char t_iobc; /* input character */
230#define TCPOOB_HAVEDATA 0x01
231#define TCPOOB_HADDATA 0x02
8acdb67c 232
984263bc
MD
233/* RFC 1323 variables */
234 u_char snd_scale; /* window scaling for send window */
235 u_char rcv_scale; /* window scaling for recv window */
236 u_char request_r_scale; /* pending window scaling */
237 u_char requested_s_scale;
238 u_long ts_recent; /* timestamp echo data */
239
240 u_long ts_recent_age; /* when last updated */
241 tcp_seq last_ack_sent;
8acdb67c 242
984263bc
MD
243/* experimental */
244 u_long snd_cwnd_prev; /* cwnd prior to retransmit */
8acdb67c 245 u_long snd_wacked_prev; /* prior bytes acked in send window */
984263bc 246 u_long snd_ssthresh_prev; /* ssthresh prior to retransmit */
9845754e 247 tcp_seq snd_recover_prev; /* snd_recover prior to retransmit */
984263bc 248 u_long t_badrxtwin; /* window for retransmit recovery */
efd4b327 249 u_long t_rexmtTS; /* timestamp of last retransmit */
33abdd1c 250 u_char snd_limited; /* segments limited transmitted */
91489f6b
JH
251
252 tcp_seq rexmt_high; /* highest seq # retransmitted + 1 */
253 tcp_seq snd_max_rexmt; /* snd_max when rexmting snd_una */
254 struct scoreboard scb; /* sack scoreboard */
255 struct raw_sackblock reportblk; /* incoming segment or D-SACK block */
256 struct raw_sackblock encloseblk;
257 int nsackhistory;
258 struct raw_sackblock sackhistory[MAX_SACK_REPORT_BLOCKS]; /* reported */
2b1ce38a 259 TAILQ_ENTRY(tcpcb) t_outputq; /* tcp_output needed list */
8acdb67c
JH
260
261 /* bandwith limitation */
262 u_long snd_bandwidth; /* calculated bandwidth or 0 */
263 u_long snd_bwnd; /* bandwidth-controlled window */
264 int t_bw_rtttime; /* used for bandwidth calculation */
265 tcp_seq t_bw_rtseq; /* used for bandwidth calculation */
5b0b9fa5
PA
266
267/* anti DoS counters */
268 u_long rcv_second; /* start of interval second */
269 u_long rcv_pps; /* received packets per second */
270 u_long rcv_byps; /* received bytes per second */
271
272 u_int32_t rfbuf_ts; /* recv buffer autoscaling timestamp */
273 int rfbuf_cnt; /* recv buffer autoscaling byte count */
2ce132be 274
340ba84c 275 int t_keepinit; /* time to establish connection */
7ea3a353 276
340ba84c
SZ
277 int t_keepidle; /* time before keepalive probes begin */
278 int t_keepintvl; /* time between keepalive probes */
279 int t_keepcnt; /* maximum number of keepalive probes */
280 int t_maxidle; /* time to drop after starting probes */
d5082e3d
SZ
281
282 int t_rxtsyn; /* time spent in SYN or SYN|ACK rexmt */
984263bc
MD
283};
284
9845754e
MD
285#define IN_FASTRECOVERY(tp) (tp->t_flags & TF_FASTRECOVERY)
286#define ENTER_FASTRECOVERY(tp) tp->t_flags |= TF_FASTRECOVERY
287#define EXIT_FASTRECOVERY(tp) tp->t_flags &= ~TF_FASTRECOVERY
288
b1992928
MD
289#ifdef TCP_SIGNATURE
290/*
291 * Defines which are needed by the xform_tcp module and tcp_[in|out]put
292 * for SADB verification and lookup.
293 */
294#define TCP_SIGLEN 16 /* length of computed digest in bytes */
295#define TCP_KEYLEN_MIN 1 /* minimum length of TCP-MD5 key */
296#define TCP_KEYLEN_MAX 80 /* maximum length of TCP-MD5 key */
297/*
298 * Only a single SA per host may be specified at this time. An SPI is
299 * needed in order for the KEY_ALLOCSA() lookup to work.
300 */
301#define TCP_SIG_SPI 0x1000
302#endif /* TCP_SIGNATURE */
303
49309f25
HP
304/*
305 * TCP statistics.
306 */
307struct tcp_stats {
308 u_long tcps_connattempt; /* connections initiated */
309 u_long tcps_accepts; /* connections accepted */
310 u_long tcps_connects; /* connections established */
311 u_long tcps_drops; /* connections dropped */
312 u_long tcps_conndrops; /* embryonic connections dropped */
5b0b9fa5 313 u_long tcps_minmssdrops; /* average minmss too low drops */
49309f25
HP
314 u_long tcps_closed; /* conn. closed (includes drops) */
315 u_long tcps_segstimed; /* segs where we tried to get rtt */
316 u_long tcps_rttupdated; /* times we succeeded */
317 u_long tcps_delack; /* delayed acks sent */
318 u_long tcps_timeoutdrop; /* conn. dropped in rxmt timeout */
319 u_long tcps_rexmttimeo; /* retransmit timeouts */
320 u_long tcps_persisttimeo; /* persist timeouts */
321 u_long tcps_keeptimeo; /* keepalive timeouts */
322 u_long tcps_keepprobe; /* keepalive probes sent */
323 u_long tcps_keepdrops; /* connections dropped in keepalive */
324
325 u_long tcps_sndtotal; /* total packets sent */
326 u_long tcps_sndpack; /* data packets sent */
327 u_long tcps_sndbyte; /* data bytes sent */
328 u_long tcps_sndrexmitpack; /* data packets retransmitted */
329 u_long tcps_sndrexmitbyte; /* data bytes retransmitted */
02cc2f35
SZ
330 u_long tcps_sndsackrtopack; /* packets sent by SACK after RTO */
331 u_long tcps_sndsackrtobyte; /* bytes sent by SACK after RTO */
49309f25
HP
332 u_long tcps_sndfastrexmit; /* Fast Retransmissions */
333 u_long tcps_sndearlyrexmit; /* early Fast Retransmissions */
334 u_long tcps_sndlimited; /* Limited Transmit packets */
335 u_long tcps_sndrtobad; /* spurious RTO retransmissions */
336 u_long tcps_sndfastrexmitbad; /* spurious Fast Retransmissions */
337 u_long tcps_sndearlyrexmitbad; /* spurious early Fast Retransmissions,
338 a subset of tcps_sndfastrexmitbad */
339 u_long tcps_eifeldetected; /* Eifel-detected spurious rexmits */
340 u_long tcps_rttcantdetect; /* Eifel but not 1/2 RTT-detectable */
341 u_long tcps_rttdetected; /* RTT-detected spurious RTO rexmits */
342 u_long tcps_sndacks; /* ack-only packets sent */
343 u_long tcps_sndprobe; /* window probes sent */
344 u_long tcps_sndurg; /* packets sent with URG only */
345 u_long tcps_sndwinup; /* window update-only packets sent */
346 u_long tcps_sndctrl; /* control (SYN|FIN|RST) packets sent */
91489f6b
JH
347 u_long tcps_sndsackpack; /* packets sent by SACK recovery alg */
348 u_long tcps_sndsackbyte; /* bytes sent by SACK recovery */
349 u_long tcps_snduna3; /* re-retransmit snd_una on 3 new seg */
350 u_long tcps_snduna1; /* re-retransmit snd_una on 1 new seg */
02cc2f35
SZ
351 u_long tcps_sndsackopt; /* SACK options sent */
352 u_long tcps_snddsackopt; /* D-SACK options sent */
3d127502 353 u_long tcps_sndidle; /* sending idle detected */
49309f25
HP
354
355 u_long tcps_rcvtotal; /* total packets received */
356 u_long tcps_rcvpack; /* packets received in sequence */
357 u_long tcps_rcvbyte; /* bytes received in sequence */
358 u_long tcps_rcvbadsum; /* packets received with ccksum errs */
359 u_long tcps_rcvbadoff; /* packets received with bad offset */
360 u_long tcps_rcvmemdrop; /* packets dropped for lack of memory */
361 u_long tcps_rcvshort; /* packets received too short */
362 u_long tcps_rcvduppack; /* duplicate-only packets received */
363 u_long tcps_rcvdupbyte; /* duplicate-only bytes received */
364 u_long tcps_rcvpartduppack; /* packets with some duplicate data */
365 u_long tcps_rcvpartdupbyte; /* dup. bytes in part-dup. packets */
366 u_long tcps_rcvoopack; /* out-of-order packets received */
367 u_long tcps_rcvoobyte; /* out-of-order bytes received */
368 u_long tcps_rcvpackafterwin; /* packets with data after window */
369 u_long tcps_rcvbyteafterwin; /* bytes rcvd after window */
370 u_long tcps_rcvafterclose; /* packets rcvd after "close" */
371 u_long tcps_rcvwinprobe; /* rcvd window probe packets */
372 u_long tcps_rcvdupack; /* rcvd duplicate acks */
373 u_long tcps_rcvacktoomuch; /* rcvd acks for unsent data */
374 u_long tcps_rcvackpack; /* rcvd ack packets */
375 u_long tcps_rcvackbyte; /* bytes acked by rcvd acks */
376 u_long tcps_rcvwinupd; /* rcvd window update packets */
377 u_long tcps_pawsdrop; /* segments dropped due to PAWS */
378 u_long tcps_predack; /* times hdr predict ok for acks */
379 u_long tcps_preddat; /* times hdr predict ok for data pkts */
380 u_long tcps_pcbcachemiss;
381 u_long tcps_cachedrtt; /* times cached RTT in route updated */
382 u_long tcps_cachedrttvar; /* times cached rttvar updated */
383 u_long tcps_cachedssthresh; /* times cached ssthresh updated */
384 u_long tcps_usedrtt; /* times RTT initialized from route */
385 u_long tcps_usedrttvar; /* times RTTVAR initialized from rt */
386 u_long tcps_usedssthresh; /* times ssthresh initialized from rt*/
387 u_long tcps_persistdrop; /* timeout in persist state */
388 u_long tcps_badsyn; /* bogus SYN, e.g. premature ACK */
389 u_long tcps_mturesent; /* resends due to MTU discovery */
390 u_long tcps_listendrop; /* listen queue overflows */
02cc2f35
SZ
391 u_long tcps_rcvbadsackopt; /* rcvd illegal SACK options */
392
393 u_long tcps_sacksbupdate; /* times SACK scoreboard updated */
394 u_long tcps_sacksboverflow; /* times SACK scoreboard overflowed */
395 u_long tcps_sacksbreused; /* times SACK sb-block reused */
396 u_long tcps_sacksbfailed; /* times SACK sb update failed */
006af0db 397 u_long tcps_sacksbfast; /* timee SACK sb-block uses cache */
49309f25
HP
398
399 u_long tcps_sc_added; /* entry added to syncache */
400 u_long tcps_sc_retransmitted; /* syncache entry was retransmitted */
401 u_long tcps_sc_dupsyn; /* duplicate SYN packet */
402 u_long tcps_sc_dropped; /* could not reply to packet */
403 u_long tcps_sc_completed; /* successful extraction of entry */
404 u_long tcps_sc_bucketoverflow; /* syncache per-bucket limit hit */
405 u_long tcps_sc_cacheoverflow; /* syncache cache limit hit */
406 u_long tcps_sc_reset; /* RST removed entry from syncache */
407 u_long tcps_sc_stale; /* timed out or listen socket gone */
408 u_long tcps_sc_aborted; /* syncache entry aborted */
409 u_long tcps_sc_badack; /* removed due to bad ACK */
410 u_long tcps_sc_unreach; /* ICMP unreachable received */
411 u_long tcps_sc_zonefail; /* zalloc() failed */
412 u_long tcps_sc_sendcookie; /* SYN cookie sent */
413 u_long tcps_sc_recvcookie; /* SYN cookie received */
414};
415
70068ffd
SW
416#ifdef _KERNEL
417
418#if defined(SMP)
419#define tcpstat tcpstats_percpu[mycpuid]
420#else
421#define tcpstat tcpstats_percpu[0]
422#endif
423
424struct sockopt;
425
426extern struct tcp_stats tcpstats_percpu[MAXCPU];
427
428static const int tcprexmtthresh = 3;
429#endif
430
984263bc
MD
431/*
432 * Structure to hold TCP options that are only used during segment
433 * processing (in tcp_input), but not held in the tcpcb.
434 * It's basically used to reduce the number of parameters
435 * to tcp_dooptions.
436 */
437struct tcpopt {
438 u_long to_flags; /* which options are present */
91489f6b 439#define TOF_TS 0x0001 /* timestamp */
91489f6b
JH
440#define TOF_MSS 0x0010
441#define TOF_SCALE 0x0020
442#define TOF_SACK_PERMITTED 0x0040
443#define TOF_SACK 0x0080
b1992928
MD
444#define TOF_SIGNATURE 0x0100 /* signature option present */
445#define TOF_SIGLEN 0x0200 /* sigature length valid (RFC2385) */
984263bc
MD
446 u_int32_t to_tsval;
447 u_int32_t to_tsecr;
984263bc 448 u_int16_t to_mss;
f23061d4 449 u_int8_t to_requested_s_scale;
16fb1aa4 450 u_int8_t to_nsackblocks;
91489f6b 451 struct raw_sackblock *to_sackblocks;
984263bc
MD
452};
453
454struct syncache {
455 inp_gen_t sc_inp_gencnt; /* pointer check */
f23061d4 456 struct tcpcb *sc_tp; /* tcb for listening socket */
984263bc 457 struct mbuf *sc_ipopts; /* source route */
f23061d4 458 struct in_conninfo sc_inc; /* addresses */
984263bc
MD
459#define sc_route sc_inc.inc_route
460#define sc_route6 sc_inc.inc6_route
461 u_int32_t sc_tsrecent;
f23061d4
JH
462 tcp_seq sc_irs; /* seq from peer */
463 tcp_seq sc_iss; /* our ISS */
984263bc 464 u_long sc_rxttime; /* retransmit time */
f23061d4 465 u_int16_t sc_rxtslot; /* retransmit counter */
984263bc
MD
466 u_int16_t sc_peer_mss; /* peer's MSS */
467 u_int16_t sc_wnd; /* advertised window */
f23061d4 468 u_int8_t sc_requested_s_scale:4,
984263bc
MD
469 sc_request_r_scale:4;
470 u_int8_t sc_flags;
a5263048
JH
471#define SCF_NOOPT 0x01 /* no TCP options */
472#define SCF_WINSCALE 0x02 /* negotiated window scaling */
473#define SCF_TIMESTAMP 0x04 /* negotiated timestamps */
d5082e3d 474#define SCF_UNUSED 0x08 /* unused */
a5263048
JH
475#define SCF_UNREACH 0x10 /* icmp unreachable received */
476#define SCF_SACK_PERMITTED 0x20 /* saw SACK permitted option */
be34e534 477#define SCF_SIGNATURE 0x40 /* send MD5 digests */
c1d0893d 478#define SCF_MARKER 0x80 /* not a real entry */
d5082e3d 479 int sc_rxtused; /* time spent in SYN|ACK rxt */
a5263048
JH
480 TAILQ_ENTRY(syncache) sc_hash;
481 TAILQ_ENTRY(syncache) sc_timerq;
984263bc
MD
482};
483
484struct syncache_head {
485 TAILQ_HEAD(, syncache) sch_bucket;
486 u_int sch_length;
487};
f23061d4 488
984263bc
MD
489#define intotcpcb(ip) ((struct tcpcb *)(ip)->inp_ppcb)
490#define sototcpcb(so) (intotcpcb(sotoinpcb(so)))
491
492/*
493 * The smoothed round-trip time and estimated variance
494 * are stored as fixed point numbers scaled by the values below.
495 * For convenience, these scales are also used in smoothing the average
496 * (smoothed = (1/scale)sample + ((scale-1)/scale)smoothed).
497 * With these scales, srtt has 3 bits to the right of the binary point,
498 * and thus an "ALPHA" of 0.875. rttvar has 2 bits to the right of the
499 * binary point, and is smoothed with an ALPHA of 0.75.
500 */
501#define TCP_RTT_SCALE 32 /* multiplier for srtt; 3 bits frac. */
502#define TCP_RTT_SHIFT 5 /* shift for srtt; 3 bits frac. */
503#define TCP_RTTVAR_SCALE 16 /* multiplier for rttvar; 2 bits */
504#define TCP_RTTVAR_SHIFT 4 /* shift for rttvar; 2 bits */
505#define TCP_DELTA_SHIFT 2 /* see tcp_input.c */
506
507/*
508 * The initial retransmission should happen at rtt + 4 * rttvar.
509 * Because of the way we do the smoothing, srtt and rttvar
510 * will each average +1/2 tick of bias. When we compute
511 * the retransmit timer, we want 1/2 tick of rounding and
512 * 1 extra tick because of +-1/2 tick uncertainty in the
513 * firing of the timer. The bias will give us exactly the
514 * 1.5 tick we need. But, because the bias is
515 * statistical, we have to test that we don't drop below
516 * the minimum feasible timer (which is 2 ticks).
517 * This version of the macro adapted from a paper by Lawrence
518 * Brakmo and Larry Peterson which outlines a problem caused
519 * by insufficient precision in the original implementation,
520 * which results in inappropriately large RTO values for very
521 * fast networks.
522 */
523#define TCP_REXMTVAL(tp) \
524 max((tp)->t_rttmin, (((tp)->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)) \
525 + (tp)->t_rttvar) >> TCP_DELTA_SHIFT)
526
527/*
984263bc
MD
528 * TCB structure exported to user-land via sysctl(3).
529 * Evil hack: declare only if in_pcb.h and sys/socketvar.h have been
530 * included. Not all of our clients do.
531 */
532#if defined(_NETINET_IN_PCB_H_) && defined(_SYS_SOCKETVAR_H_)
533struct xtcpcb {
534 size_t xt_len;
535 struct inpcb xt_inp;
536 struct tcpcb xt_tp;
537 struct xsocket xt_socket;
538 u_quad_t xt_alignment_hack;
539};
540#endif
541
542/*
543 * Names for TCP sysctl objects
544 */
545#define TCPCTL_DO_RFC1323 1 /* use RFC-1323 extensions */
27b8aee3 546/* 2 was TCPCTL_DO_RFC1644 */
984263bc
MD
547#define TCPCTL_MSSDFLT 3 /* MSS default */
548#define TCPCTL_STATS 4 /* statistics (read-only) */
549#define TCPCTL_RTTDFLT 5 /* default RTT estimate */
550#define TCPCTL_KEEPIDLE 6 /* keepalive idle timer */
551#define TCPCTL_KEEPINTVL 7 /* interval to send keepalives */
552#define TCPCTL_SENDSPACE 8 /* send buffer space */
553#define TCPCTL_RECVSPACE 9 /* receive buffer space */
554#define TCPCTL_KEEPINIT 10 /* timeout for establishing syn */
555#define TCPCTL_PCBLIST 11 /* list of all outstanding PCBs */
556#define TCPCTL_DELACKTIME 12 /* time before sending delayed ACK */
557#define TCPCTL_V6MSSDFLT 13 /* MSS default for IPv6 */
558#define TCPCTL_MAXID 14
559
560#define TCPCTL_NAMES { \
561 { 0, 0 }, \
562 { "rfc1323", CTLTYPE_INT }, \
27b8aee3 563 { "reserved", CTLTYPE_INT}, /* was rfc1644 */ \
984263bc
MD
564 { "mssdflt", CTLTYPE_INT }, \
565 { "stats", CTLTYPE_STRUCT }, \
566 { "rttdflt", CTLTYPE_INT }, \
567 { "keepidle", CTLTYPE_INT }, \
568 { "keepintvl", CTLTYPE_INT }, \
569 { "sendspace", CTLTYPE_INT }, \
570 { "recvspace", CTLTYPE_INT }, \
571 { "keepinit", CTLTYPE_INT }, \
572 { "pcblist", CTLTYPE_STRUCT }, \
573 { "delacktime", CTLTYPE_INT }, \
574 { "v6mssdflt", CTLTYPE_INT }, \
575}
576
984263bc
MD
577#ifdef _KERNEL
578#ifdef SYSCTL_DECL
579SYSCTL_DECL(_net_inet_tcp);
580#endif
581
91489f6b 582#define TCP_DO_SACK(tp) ((tp)->t_flags & TF_SACK_PERMITTED)
3a5d999b
SZ
583#define TCP_SACK_BLKEND(len, thflags) \
584 ((len) + (((thflags) & TH_FIN) != 0))
91489f6b 585
2b1ce38a
MD
586TAILQ_HEAD(tcpcbackqhead,tcpcb);
587
d371a63a 588extern struct inpcbinfo tcbinfo[];
2b1ce38a
MD
589extern struct tcpcbackqhead tcpcbackq[];
590
984263bc 591extern int tcp_mssdflt; /* XXX */
5b0b9fa5 592extern int tcp_minmss;
984263bc 593extern int tcp_delack_enabled;
984263bc 594extern int path_mtu_discovery;
984263bc 595
002c1265
MD
596union netmsg;
597
d371a63a
JH
598int tcp_addrcpu(in_addr_t faddr, in_port_t fport,
599 in_addr_t laddr, in_port_t lport);
7fe56515
JH
600struct lwkt_port *
601 tcp_addrport(in_addr_t faddr, in_port_t fport,
602 in_addr_t laddr, in_port_t lport);
65f3e756 603struct lwkt_port *tcp_addrport0(void);
42a7fc75 604void tcp_canceltimers (struct tcpcb *);
984263bc 605struct tcpcb *
42a7fc75 606 tcp_close (struct tcpcb *);
002c1265
MD
607void tcp_ctlinput(union netmsg *);
608void tcp_ctloutput(union netmsg *);
984263bc 609struct tcpcb *
42a7fc75
RG
610 tcp_drop (struct tcpcb *, int);
611void tcp_drain (void);
612void tcp_fasttimo (void);
42a7fc75 613void tcp_init (void);
bf82f9b7 614void tcp_thread_init (void);
002c1265 615int tcp_input (struct mbuf **, int *, int);
42a7fc75
RG
616void tcp_mss (struct tcpcb *, int);
617int tcp_mssopt (struct tcpcb *);
618void tcp_drop_syn_sent (struct inpcb *, int);
619void tcp_mtudisc (struct inpcb *, int);
984263bc 620struct tcpcb *
42a7fc75
RG
621 tcp_newtcpcb (struct inpcb *);
622int tcp_output (struct tcpcb *);
623void tcp_quench (struct inpcb *, int);
624void tcp_respond (struct tcpcb *, void *,
625 struct tcphdr *, struct mbuf *, tcp_seq, tcp_seq, int);
984263bc 626struct rtentry *
42a7fc75 627 tcp_rtlookup (struct in_conninfo *);
9a5b142f 628int tcp_sack_bytes_below(struct scoreboard *scb, tcp_seq seq);
006af0db 629void tcp_sack_destroy(struct scoreboard *scb);
91489f6b 630void tcp_sack_cleanup(struct scoreboard *scb);
d58ca578 631void tcp_sack_report_cleanup(struct tcpcb *tp);
91489f6b
JH
632int tcp_sack_ndsack_blocks(struct raw_sackblock *blocks,
633 const int numblocks, tcp_seq snd_una);
634void tcp_sack_fill_report(struct tcpcb *tp, u_char *opt, u_int *plen);
635boolean_t
636 tcp_sack_has_sacked(struct scoreboard *scb, u_int amount);
91489f6b
JH
637void tcp_sack_tcpcb_init(struct tcpcb *tp);
638uint32_t tcp_sack_compute_pipe(struct tcpcb *tp);
639boolean_t
640 tcp_sack_nextseg(struct tcpcb *tp, tcp_seq *nextrexmt, uint32_t *len,
641 boolean_t *losdup);
642#ifdef later
643void tcp_sack_revert_scoreboard(struct scoreboard *scb, tcp_seq snd_una,
644 u_int maxseg);
645void tcp_sack_save_scoreboard(struct scoreboard *scb);
646#endif
647void tcp_sack_skip_sacked(struct scoreboard *scb, tcp_seq *prexmt);
648void tcp_sack_update_scoreboard(struct tcpcb *tp, struct tcpopt *to);
efd4b327 649void tcp_save_congestion_state(struct tcpcb *tp);
bfdb979e 650void tcp_revert_congestion_state(struct tcpcb *tp);
42a7fc75 651void tcp_setpersist (struct tcpcb *);
dd2b0fb4
MD
652struct tcptemp *tcp_maketemplate (struct tcpcb *);
653void tcp_freetemplate (struct tcptemp *);
42a7fc75 654void tcp_fillheaders (struct tcpcb *, void *, void *);
bf82f9b7 655struct lwkt_port *
48e7b118
MD
656 tcp_soport(struct socket *, struct sockaddr *, struct mbuf **);
657struct lwkt_port *
e3873585 658 tcp_ctlport(int, struct sockaddr *, void *);
984263bc 659struct tcpcb *
42a7fc75 660 tcp_timers (struct tcpcb *, int);
f3f70f0d 661void tcp_trace (short, short, struct tcpcb *, void *, struct tcphdr *,
42a7fc75 662 int);
984263bc 663void tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq);
c01148a3 664u_long tcp_initial_window(struct tcpcb *tp);
0ecd93f9 665void tcp_timer_keep_activity(struct tcpcb *tp, int thflags);
984263bc
MD
666void syncache_init(void);
667void syncache_unreach(struct in_conninfo *, struct tcphdr *);
668int syncache_expand(struct in_conninfo *, struct tcphdr *,
669 struct socket **, struct mbuf *);
670int syncache_add(struct in_conninfo *, struct tcpopt *,
b09567cc 671 struct tcphdr *, struct socket *, struct mbuf *);
984263bc
MD
672void syncache_chkrst(struct in_conninfo *, struct tcphdr *);
673void syncache_badack(struct in_conninfo *);
e5fe3477
MD
674void syncache_destroy(struct tcpcb *tp);
675
b1992928
MD
676#ifdef TCP_SIGNATURE
677int tcpsignature_apply(void *fstate, void *data, unsigned int len);
678int tcpsignature_compute(struct mbuf *m, int len, int tcpoptlen,
679 u_char *buf, u_int direction);
680#endif /* TCP_SIGNATURE */
984263bc
MD
681
682extern struct pr_usrreqs tcp_usrreqs;
683extern u_long tcp_sendspace;
684extern u_long tcp_recvspace;
42a7fc75 685tcp_seq tcp_new_isn (struct tcpcb *);
984263bc
MD
686
687#endif /* _KERNEL */
688
689#endif /* _NETINET_TCP_VAR_H_ */