from van: header prediction, pcb one-element cache, avoid pullup,
remove dtom, avoid spurious assignment from null ntoh*; faster window opening
in linear mode (add 1/8 packet, was 146 bytes), open w

from van: header prediction, pcb one-element cache, avoid pullup,
remove dtom, avoid spurious assignment from null ntoh*; faster window opening
in linear mode (add 1/8 packet, was 146 bytes), open window past cached
out-of order packets until dupacks end; rearrange timers; use rtentry parms

show more ...

sonewconn has second arg; in_pcb.h requires ip.h

sockaddr lens

update branch to 7.19

mark state of kernel before variable length sockaddrs (by sklower).

install approved copyright notice

ack immediately when segment is received out of order;
remove tcpnodelack

update branch

much longer keepalive timer (don't start probes for 2 hr);
fix input window test again (accept acks from FINS at left edge and responses
to keepalives, but dropafterack otherwise; also fixes Phil Kar

much longer keepalive timer (don't start probes for 2 hr);
fix input window test again (accept acks from FINS at left edge and responses
to keepalives, but dropafterack otherwise; also fixes Phil Karn's problem
with failure to respond to new SYN to left of window); can't use tcp_drop
after RST, or we'll send RST in response

show more ...

65535 => IP_MAXPACKET (still XXX)

branch for FTP release (#ifdefs for non-4.3)

ack pure window probes; process URG, etc. even if window is closed;
squish out common code for window open/closed cases

use Berkeley specific header

window isn't necessarily zero when user is gone, don't accept new data

slow down more when doing fast retransmission

two-phase slow start; fast retransmit on repetitive ack;
drop packet for TCP socket between socket() and listen(); get the rtt
on SYN

Phil Karn's retransmit algorithm, using t_rxtcur for current rexmt value;
slow start from beginning if connection not local; get rtt estimate on SYN

free those mbufs! (and skip redundant trace)

we go to LAST_ACK even if data is outstanding, so watch for window
updates and acks in LAST_ACK

respond to those resets the keepalives produce

redo retransmit calculations as srtt + 2*smoothed variance
(from Van Jacobson)

cleanups: correct snd_una and timer handling when SYN acked
from SYN_RCVD state, so rtt is collected; use tcp_output instead of tcp_respond
to ack (may need to send FIN, etc); clarify obscure code (o

cleanups: correct snd_una and timer handling when SYN acked
from SYN_RCVD state, so rtt is collected; use tcp_output instead of tcp_respond
to ack (may need to send FIN, etc); clarify obscure code (ourfinisacked)

show more ...

slow-start changes from van

don't assume ACK on received SYN when in SYN_SENT
(and don't loop when a socket is connected to itself)

new stats; rearrange to trim data from front before noting
that data is received after user is gone (don't reset because of dup data!);
quick connection reuse