freebsd-nq/sys/netinet/tcp_var.h

492 lines
19 KiB
C
Raw Normal View History

1994-05-24 10:09:53 +00:00
/*
* Copyright (c) 1982, 1986, 1993, 1994, 1995
1994-05-24 10:09:53 +00:00
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)tcp_var.h 8.4 (Berkeley) 5/24/95
1999-08-28 01:08:13 +00:00
* $FreeBSD$
1994-05-24 10:09:53 +00:00
*/
1994-08-21 05:27:42 +00:00
#ifndef _NETINET_TCP_VAR_H_
#define _NETINET_TCP_VAR_H_
#include <netinet/in_pcb.h> /* needed for in_conninfo, inp_gen_t */
#include <netinet/tcp.h>
1994-05-24 10:09:53 +00:00
/*
* Kernel variables for tcp.
*/
extern int tcp_do_rfc1323;
extern int tcp_do_rfc1644;
1994-05-24 10:09:53 +00:00
/* TCP segment queue entry */
struct tseg_qent {
LIST_ENTRY(tseg_qent) tqe_q;
int tqe_len; /* TCP segment data length */
struct tcphdr *tqe_th; /* a pointer to tcp header */
struct mbuf *tqe_m; /* mbuf contains packet */
};
LIST_HEAD(tsegqe_head, tseg_qent);
#ifdef MALLOC_DECLARE
MALLOC_DECLARE(M_TSEGQ);
#endif
struct tcptemp {
u_char tt_ipgen[40]; /* the size must be of max ip header, now IPv6 */
struct tcphdr tt_t;
};
#define tcp6cb tcpcb /* for KAME src sync over BSD*'s */
1994-05-24 10:09:53 +00:00
/*
* Tcp control block, one per tcp; fields:
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
* Organized for 16 byte cacheline efficiency.
1994-05-24 10:09:53 +00:00
*/
struct tcpcb {
struct tsegqe_head t_segq;
int t_dupacks; /* consecutive dup acks recd */
struct tcptemp *unused; /* unused */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
struct callout *tt_rexmt; /* retransmit timer */
struct callout *tt_persist; /* retransmit persistence */
struct callout *tt_keep; /* keepalive */
struct callout *tt_2msl; /* 2*msl TIME_WAIT timer */
struct callout *tt_delack; /* delayed ACK timer */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
struct inpcb *t_inpcb; /* back pointer to internet pcb */
int t_state; /* state of this connection */
u_int t_flags;
#define TF_ACKNOW 0x00001 /* ack peer immediately */
#define TF_DELACK 0x00002 /* ack, but try to delay it */
#define TF_NODELAY 0x00004 /* don't delay packets to coalesce */
#define TF_NOOPT 0x00008 /* don't use tcp options */
#define TF_SENTFIN 0x00010 /* have sent FIN */
#define TF_REQ_SCALE 0x00020 /* have/will request window scaling */
#define TF_RCVD_SCALE 0x00040 /* other side has requested scaling */
#define TF_REQ_TSTMP 0x00080 /* have/will request timestamps */
#define TF_RCVD_TSTMP 0x00100 /* a timestamp was received in SYN */
#define TF_SACK_PERMIT 0x00200 /* other side said I could SACK */
#define TF_NEEDSYN 0x00400 /* send SYN (implicit state) */
#define TF_NEEDFIN 0x00800 /* send FIN (implicit state) */
#define TF_NOPUSH 0x01000 /* don't push */
#define TF_REQ_CC 0x02000 /* have/will request CC */
#define TF_RCVD_CC 0x04000 /* a CC was received in SYN */
#define TF_SENDCCNEW 0x08000 /* send CCnew instead of CC in SYN */
#define TF_MORETOCOME 0x10000 /* More data to be appended to sock */
#define TF_LQ_OVERFLOW 0x20000 /* listen queue overflow */
#define TF_LASTIDLE 0x40000 /* connection was previously idle */
#define TF_RXWIN0SENT 0x80000 /* sent a receiver win 0 in response */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
int t_force; /* 1 if forcing out a byte */
1994-05-24 10:09:53 +00:00
tcp_seq snd_una; /* send unacknowledged */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
tcp_seq snd_max; /* highest sequence number sent;
* used to recognize retransmits
*/
1994-05-24 10:09:53 +00:00
tcp_seq snd_nxt; /* send next */
tcp_seq snd_up; /* send urgent pointer */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
1994-05-24 10:09:53 +00:00
tcp_seq snd_wl1; /* window update seg seq number */
tcp_seq snd_wl2; /* window update seg ack number */
tcp_seq iss; /* initial send sequence number */
tcp_seq irs; /* initial receive sequence number */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
tcp_seq rcv_nxt; /* receive next */
1994-05-24 10:09:53 +00:00
tcp_seq rcv_adv; /* advertised window */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
u_long rcv_wnd; /* receive window */
tcp_seq rcv_up; /* receive urgent pointer */
u_long snd_wnd; /* send window */
1994-05-24 10:09:53 +00:00
u_long snd_cwnd; /* congestion-controlled window */
u_long snd_ssthresh; /* snd_cwnd size threshold for
1994-05-24 10:09:53 +00:00
* for slow start exponential to
* linear switch
*/
tcp_seq snd_recover; /* for use in fast recovery */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
u_int t_maxopd; /* mss plus options */
u_long t_rcvtime; /* inactivity time */
u_long t_starttime; /* time connection was established */
int t_rtttime; /* round trip time */
1994-05-24 10:09:53 +00:00
tcp_seq t_rtseq; /* sequence number being timed */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
int t_rxtcur; /* current retransmit value (ticks) */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
u_int t_maxseg; /* maximum segment size */
int t_srtt; /* smoothed round-trip time */
int t_rttvar; /* variance in round-trip time */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
int t_rxtshift; /* log(2) of rexmt exp. backoff */
u_int t_rttmin; /* minimum rtt allowed */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
u_long t_rttupdated; /* number of times rtt sampled */
1994-05-24 10:09:53 +00:00
u_long max_sndwnd; /* largest window peer has offered */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
int t_softerror; /* possible error not yet reported */
1994-05-24 10:09:53 +00:00
/* out-of-band data */
char t_oobflags; /* have some */
char t_iobc; /* input character */
#define TCPOOB_HAVEDATA 0x01
#define TCPOOB_HADDATA 0x02
/* RFC 1323 variables */
u_char snd_scale; /* window scaling for send window */
u_char rcv_scale; /* window scaling for recv window */
u_char request_r_scale; /* pending window scaling */
u_char requested_s_scale;
u_long ts_recent; /* timestamp echo data */
Improved connection establishment performance by doing local port lookups via a hashed port list. In the new scheme, in_pcblookup() goes away and is replaced by a new routine, in_pcblookup_local() for doing the local port check. Note that this implementation is space inefficient in that the PCB struct is now too large to fit into 128 bytes. I might deal with this in the future by using the new zone allocator, but I wanted these changes to be extensively tested in their current form first. Also: 1) Fixed off-by-one errors in the port lookup loops in in_pcbbind(). 2) Got rid of some unneeded rehashing. Adding a new routine, in_pcbinshash() to do the initialial hash insertion. 3) Renamed in_pcblookuphash() to in_pcblookup_hash() for easier readability. 4) Added a new routine, in_pcbremlists() to remove the PCB from the various hash lists. 5) Added/deleted comments where appropriate. 6) Removed unnecessary splnet() locking. In general, the PCB functions should be called at splnet()...there are unfortunately a few exceptions, however. 7) Reorganized a few structs for better cache line behavior. 8) Killed my TCP_ACK_HACK kludge. It may come back in a different form in the future, however. These changes have been tested on wcarchive for more than a month. In tests done here, connection establishment overhead is reduced by more than 50 times, thus getting rid of one of the major networking scalability problems. Still to do: make tcp_fastimo/tcp_slowtimo scale well for systems with a large number of connections. tcp_fastimo is easy; tcp_slowtimo is difficult. WARNING: Anything that knows about inpcb and tcpcb structs will have to be recompiled; at the very least, this includes netstat(1).
1998-01-27 09:15:13 +00:00
1994-05-24 10:09:53 +00:00
u_long ts_recent_age; /* when last updated */
tcp_seq last_ack_sent;
/* RFC 1644 variables */
tcp_cc cc_send; /* send connection count */
tcp_cc cc_recv; /* receive connection count */
/* experimental */
u_long snd_cwnd_prev; /* cwnd prior to retransmit */
u_long snd_ssthresh_prev; /* ssthresh prior to retransmit */
u_long t_badrxtwin; /* window for retransmit recovery */
1994-05-24 10:09:53 +00:00
};
/*
* Structure to hold TCP options that are only used during segment
* processing (in tcp_input), but not held in the tcpcb.
* It's basically used to reduce the number of parameters
* to tcp_dooptions.
*/
struct tcpopt {
u_long to_flags; /* which options are present */
#define TOF_TS 0x0001 /* timestamp */
#define TOF_CC 0x0002 /* CC and CCnew are exclusive */
#define TOF_CCNEW 0x0004
#define TOF_CCECHO 0x0008
#define TOF_MSS 0x0010
#define TOF_SCALE 0x0020
u_int32_t to_tsval;
u_int32_t to_tsecr;
tcp_cc to_cc; /* holds CC or CCnew */
tcp_cc to_ccecho;
u_int16_t to_mss;
u_int8_t to_requested_s_scale;
u_int8_t to_pad;
};
struct syncache {
inp_gen_t sc_inp_gencnt; /* pointer check */
struct tcpcb *sc_tp; /* tcb for listening socket */
struct mbuf *sc_ipopts; /* source route */
struct in_conninfo sc_inc; /* addresses */
#define sc_route sc_inc.inc_route
#define sc_route6 sc_inc.inc6_route
u_int32_t sc_tsrecent;
tcp_cc sc_cc_send; /* holds CC or CCnew */
tcp_cc sc_cc_recv;
tcp_seq sc_irs; /* seq from peer */
tcp_seq sc_iss; /* our ISS */
u_long sc_rxttime; /* retransmit time */
u_int16_t sc_rxtslot; /* retransmit counter */
u_int16_t sc_peer_mss; /* peer's MSS */
u_int16_t sc_wnd; /* advertised window */
u_int8_t sc_requested_s_scale:4,
sc_request_r_scale:4;
u_int8_t sc_flags;
#define SCF_NOOPT 0x01 /* no TCP options */
#define SCF_WINSCALE 0x02 /* negotiated window scaling */
#define SCF_TIMESTAMP 0x04 /* negotiated timestamps */
#define SCF_CC 0x08 /* negotiated CC */
#define SCF_UNREACH 0x10 /* icmp unreachable received */
#define SCF_KEEPROUTE 0x20 /* keep cloned route */
TAILQ_ENTRY(syncache) sc_hash;
TAILQ_ENTRY(syncache) sc_timerq;
};
struct syncache_head {
TAILQ_HEAD(, syncache) sch_bucket;
u_int sch_length;
};
/*
* The TAO cache entry which is stored in the protocol family specific
* portion of the route metrics.
*/
struct rmxp_tao {
tcp_cc tao_cc; /* latest CC in valid SYN */
tcp_cc tao_ccsent; /* latest CC sent to peer */
u_short tao_mssopt; /* peer's cached MSS */
#ifdef notyet
u_short tao_flags; /* cache status flags */
#define TAOF_DONT 0x0001 /* peer doesn't understand rfc1644 */
#define TAOF_OK 0x0002 /* peer does understand rfc1644 */
#define TAOF_UNDEF 0 /* we don't know yet */
#endif /* notyet */
};
#define rmx_taop(r) ((struct rmxp_tao *)(r).rmx_filler)
1994-05-24 10:09:53 +00:00
#define intotcpcb(ip) ((struct tcpcb *)(ip)->inp_ppcb)
#define sototcpcb(so) (intotcpcb(sotoinpcb(so)))
/*
* The smoothed round-trip time and estimated variance
* are stored as fixed point numbers scaled by the values below.
* For convenience, these scales are also used in smoothing the average
* (smoothed = (1/scale)sample + ((scale-1)/scale)smoothed).
* With these scales, srtt has 3 bits to the right of the binary point,
* and thus an "ALPHA" of 0.875. rttvar has 2 bits to the right of the
* binary point, and is smoothed with an ALPHA of 0.75.
*/
#define TCP_RTT_SCALE 32 /* multiplier for srtt; 3 bits frac. */
#define TCP_RTT_SHIFT 5 /* shift for srtt; 3 bits frac. */
#define TCP_RTTVAR_SCALE 16 /* multiplier for rttvar; 2 bits */
#define TCP_RTTVAR_SHIFT 4 /* shift for rttvar; 2 bits */
#define TCP_DELTA_SHIFT 2 /* see tcp_input.c */
1994-05-24 10:09:53 +00:00
/*
* The initial retransmission should happen at rtt + 4 * rttvar.
* Because of the way we do the smoothing, srtt and rttvar
* will each average +1/2 tick of bias. When we compute
* the retransmit timer, we want 1/2 tick of rounding and
* 1 extra tick because of +-1/2 tick uncertainty in the
* firing of the timer. The bias will give us exactly the
* 1.5 tick we need. But, because the bias is
* statistical, we have to test that we don't drop below
* the minimum feasible timer (which is 2 ticks).
* This version of the macro adapted from a paper by Lawrence
* Brakmo and Larry Peterson which outlines a problem caused
* by insufficient precision in the original implementation,
* which results in inappropriately large RTO values for very
* fast networks.
1994-05-24 10:09:53 +00:00
*/
#define TCP_REXMTVAL(tp) \
max((tp)->t_rttmin, (((tp)->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)) \
+ (tp)->t_rttvar) >> TCP_DELTA_SHIFT)
1994-05-24 10:09:53 +00:00
/*
* TCP statistics.
* Many of these should be kept per connection,
* but that's inconvenient at the moment.
*/
struct tcpstat {
u_long tcps_connattempt; /* connections initiated */
u_long tcps_accepts; /* connections accepted */
u_long tcps_connects; /* connections established */
u_long tcps_drops; /* connections dropped */
u_long tcps_conndrops; /* embryonic connections dropped */
u_long tcps_closed; /* conn. closed (includes drops) */
u_long tcps_segstimed; /* segs where we tried to get rtt */
u_long tcps_rttupdated; /* times we succeeded */
u_long tcps_delack; /* delayed acks sent */
u_long tcps_timeoutdrop; /* conn. dropped in rxmt timeout */
u_long tcps_rexmttimeo; /* retransmit timeouts */
u_long tcps_persisttimeo; /* persist timeouts */
u_long tcps_keeptimeo; /* keepalive timeouts */
u_long tcps_keepprobe; /* keepalive probes sent */
u_long tcps_keepdrops; /* connections dropped in keepalive */
u_long tcps_sndtotal; /* total packets sent */
u_long tcps_sndpack; /* data packets sent */
u_long tcps_sndbyte; /* data bytes sent */
u_long tcps_sndrexmitpack; /* data packets retransmitted */
u_long tcps_sndrexmitbyte; /* data bytes retransmitted */
u_long tcps_sndacks; /* ack-only packets sent */
u_long tcps_sndprobe; /* window probes sent */
u_long tcps_sndurg; /* packets sent with URG only */
u_long tcps_sndwinup; /* window update-only packets sent */
u_long tcps_sndctrl; /* control (SYN|FIN|RST) packets sent */
u_long tcps_rcvtotal; /* total packets received */
u_long tcps_rcvpack; /* packets received in sequence */
u_long tcps_rcvbyte; /* bytes received in sequence */
u_long tcps_rcvbadsum; /* packets received with ccksum errs */
u_long tcps_rcvbadoff; /* packets received with bad offset */
u_long tcps_rcvmemdrop; /* packets dropped for lack of memory */
1994-05-24 10:09:53 +00:00
u_long tcps_rcvshort; /* packets received too short */
u_long tcps_rcvduppack; /* duplicate-only packets received */
u_long tcps_rcvdupbyte; /* duplicate-only bytes received */
u_long tcps_rcvpartduppack; /* packets with some duplicate data */
u_long tcps_rcvpartdupbyte; /* dup. bytes in part-dup. packets */
u_long tcps_rcvoopack; /* out-of-order packets received */
u_long tcps_rcvoobyte; /* out-of-order bytes received */
u_long tcps_rcvpackafterwin; /* packets with data after window */
u_long tcps_rcvbyteafterwin; /* bytes rcvd after window */
u_long tcps_rcvafterclose; /* packets rcvd after "close" */
u_long tcps_rcvwinprobe; /* rcvd window probe packets */
u_long tcps_rcvdupack; /* rcvd duplicate acks */
u_long tcps_rcvacktoomuch; /* rcvd acks for unsent data */
u_long tcps_rcvackpack; /* rcvd ack packets */
u_long tcps_rcvackbyte; /* bytes acked by rcvd acks */
u_long tcps_rcvwinupd; /* rcvd window update packets */
u_long tcps_pawsdrop; /* segments dropped due to PAWS */
u_long tcps_predack; /* times hdr predict ok for acks */
u_long tcps_preddat; /* times hdr predict ok for data pkts */
u_long tcps_pcbcachemiss;
u_long tcps_cachedrtt; /* times cached RTT in route updated */
u_long tcps_cachedrttvar; /* times cached rttvar updated */
u_long tcps_cachedssthresh; /* times cached ssthresh updated */
u_long tcps_usedrtt; /* times RTT initialized from route */
u_long tcps_usedrttvar; /* times RTTVAR initialized from rt */
u_long tcps_usedssthresh; /* times ssthresh initialized from rt*/
u_long tcps_persistdrop; /* timeout in persist state */
u_long tcps_badsyn; /* bogus SYN, e.g. premature ACK */
u_long tcps_mturesent; /* resends due to MTU discovery */
u_long tcps_listendrop; /* listen queue overflows */
u_long tcps_sc_added; /* entry added to syncache */
u_long tcps_sc_retransmitted; /* syncache entry was retransmitted */
u_long tcps_sc_dupsyn; /* duplicate SYN packet */
u_long tcps_sc_dropped; /* could not reply to packet */
u_long tcps_sc_completed; /* successful extraction of entry */
u_long tcps_sc_bucketoverflow; /* syncache per-bucket limit hit */
u_long tcps_sc_cacheoverflow; /* syncache cache limit hit */
u_long tcps_sc_reset; /* RST removed entry from syncache */
u_long tcps_sc_stale; /* timed out or listen socket gone */
u_long tcps_sc_aborted; /* syncache entry aborted */
u_long tcps_sc_badack; /* removed due to bad ACK */
u_long tcps_sc_unreach; /* ICMP unreachable received */
u_long tcps_sc_zonefail; /* zalloc() failed */
u_long tcps_sc_sendcookie; /* SYN cookie sent */
u_long tcps_sc_recvcookie; /* SYN cookie received */
1994-05-24 10:09:53 +00:00
};
/*
* TCB structure exported to user-land via sysctl(3).
* Evil hack: declare only if in_pcb.h and sys/socketvar.h have been
* included. Not all of our clients do.
*/
#if defined(_NETINET_IN_PCB_H_) && defined(_SYS_SOCKETVAR_H_)
struct xtcpcb {
size_t xt_len;
struct inpcb xt_inp;
struct tcpcb xt_tp;
struct xsocket xt_socket;
u_quad_t xt_alignment_hack;
};
#endif
/*
* Names for TCP sysctl objects
*/
#define TCPCTL_DO_RFC1323 1 /* use RFC-1323 extensions */
#define TCPCTL_DO_RFC1644 2 /* use RFC-1644 extensions */
#define TCPCTL_MSSDFLT 3 /* MSS default */
#define TCPCTL_STATS 4 /* statistics (read-only) */
#define TCPCTL_RTTDFLT 5 /* default RTT estimate */
#define TCPCTL_KEEPIDLE 6 /* keepalive idle timer */
#define TCPCTL_KEEPINTVL 7 /* interval to send keepalives */
#define TCPCTL_SENDSPACE 8 /* send buffer space */
#define TCPCTL_RECVSPACE 9 /* receive buffer space */
#define TCPCTL_KEEPINIT 10 /* timeout for establishing syn */
#define TCPCTL_PCBLIST 11 /* list of all outstanding PCBs */
#define TCPCTL_DELACKTIME 12 /* time before sending delayed ACK */
#define TCPCTL_V6MSSDFLT 13 /* MSS default for IPv6 */
#define TCPCTL_MAXID 14
#define TCPCTL_NAMES { \
{ 0, 0 }, \
{ "rfc1323", CTLTYPE_INT }, \
{ "rfc1644", CTLTYPE_INT }, \
{ "mssdflt", CTLTYPE_INT }, \
{ "stats", CTLTYPE_STRUCT }, \
{ "rttdflt", CTLTYPE_INT }, \
{ "keepidle", CTLTYPE_INT }, \
{ "keepintvl", CTLTYPE_INT }, \
{ "sendspace", CTLTYPE_INT }, \
{ "recvspace", CTLTYPE_INT }, \
{ "keepinit", CTLTYPE_INT }, \
{ "pcblist", CTLTYPE_STRUCT }, \
{ "delacktime", CTLTYPE_INT }, \
{ "v6mssdflt", CTLTYPE_INT }, \
}
#ifdef _KERNEL
#ifdef SYSCTL_DECL
SYSCTL_DECL(_net_inet_tcp);
#endif
extern struct inpcbhead tcb; /* head of queue of active tcpcb's */
extern struct inpcbinfo tcbinfo;
extern struct tcpstat tcpstat; /* tcp statistics */
extern int tcp_mssdflt; /* XXX */
extern int tcp_delack_enabled;
extern int tcp_do_newreno;
extern int path_mtu_discovery;
extern int ss_fltsz;
extern int ss_fltsz_local;
1994-05-24 10:09:53 +00:00
2002-03-19 21:25:46 +00:00
void tcp_canceltimers(struct tcpcb *);
1994-05-24 10:09:53 +00:00
struct tcpcb *
2002-03-19 21:25:46 +00:00
tcp_close(struct tcpcb *);
void tcp_ctlinput(int, struct sockaddr *, void *);
int tcp_ctloutput(struct socket *, struct sockopt *);
1994-05-24 10:09:53 +00:00
struct tcpcb *
2002-03-19 21:25:46 +00:00
tcp_drop(struct tcpcb *, int);
void tcp_drain(void);
void tcp_fasttimo(void);
struct rmxp_tao *
2002-03-19 21:25:46 +00:00
tcp_gettaocache(struct in_conninfo *);
void tcp_init(void);
void tcp_input(struct mbuf *, int);
void tcp_mss(struct tcpcb *, int);
int tcp_mssopt(struct tcpcb *);
struct inpcb *
tcp_drop_syn_sent(struct inpcb *, int);
struct inpcb *
tcp_mtudisc(struct inpcb *, int);
1994-05-24 10:09:53 +00:00
struct tcpcb *
2002-03-19 21:25:46 +00:00
tcp_newtcpcb(struct inpcb *);
int tcp_output(struct tcpcb *);
struct inpcb *
tcp_quench(struct inpcb *, int);
2002-03-19 21:25:46 +00:00
void tcp_respond(struct tcpcb *, void *,
struct tcphdr *, struct mbuf *, tcp_seq, tcp_seq, int);
struct rtentry *
2002-03-19 21:25:46 +00:00
tcp_rtlookup(struct in_conninfo *);
void tcp_setpersist(struct tcpcb *);
void tcp_slowtimo(void);
struct tcptemp *
2002-03-19 21:25:46 +00:00
tcp_maketemplate(struct tcpcb *);
void tcp_fillheaders(struct tcpcb *, void *, void *);
1994-05-24 10:09:53 +00:00
struct tcpcb *
2002-03-19 21:25:46 +00:00
tcp_timers(struct tcpcb *, int);
void tcp_trace(int, int, struct tcpcb *, void *, struct tcphdr *, int);
void syncache_init(void);
void syncache_unreach(struct in_conninfo *, struct tcphdr *);
int syncache_expand(struct in_conninfo *, struct tcphdr *,
struct socket **, struct mbuf *);
int syncache_add(struct in_conninfo *, struct tcpopt *,
struct tcphdr *, struct socket **, struct mbuf *);
void syncache_chkrst(struct in_conninfo *, struct tcphdr *);
void syncache_badack(struct in_conninfo *);
extern struct pr_usrreqs tcp_usrreqs;
extern u_long tcp_sendspace;
extern u_long tcp_recvspace;
2002-03-19 21:25:46 +00:00
tcp_seq tcp_new_isn(struct tcpcb *);
#endif /* _KERNEL */
1994-08-21 05:27:42 +00:00
#endif /* _NETINET_TCP_VAR_H_ */