fix typo in tcp_random18 ifdef

change iss increment to agree with tcp_input (and tcp_seq.h comments)

from Rich Stevens, fix connection-reuse code to use send rather than
recv sequence for iss on new connection

changes to randomize iss (somewhat); use a random component in the increment.
Increase rate of periodic increment, but leave the per-connection increment
about the same. Small cleanup in code to han

changes to randomize iss (somewhat); use a random component in the increment.
Increase rate of periodic increment, but leave the per-connection increment
about the same. Small cleanup in code to handle bad (e.g. spoofed) SYN:
remove redundant code, add counter (perhaps should count only if SYN is set),
and simplify slightly for RST.

show more ...

change t_idle to unsigned (should probably be long);
two new counters

two versions of a fix for an mbuf leak. The first one is probably right,
but a BSDI customer has not confirmed that problems went away. The second
version should be safe. The mbuf leak happens whe

two versions of a fix for an mbuf leak. The first one is probably right,
but a BSDI customer has not confirmed that problems went away. The second
version should be safe. The mbuf leak happens when RST is received on
a connection after sosend checks the state, then drops spl while doing
a copyin. A send of [100-208] bytes on a local connection would always
hit this if the other end had closed.

show more ...

hack to time out connections with a zero window and a dead peer.
if the peer does not respond in the time we would allow for retransmission,
or within a fixed time, drop the connection. Minor optimi

hack to time out connections with a zero window and a dead peer.
if the peer does not respond in the time we would allow for retransmission,
or within a fixed time, drop the connection. Minor optimization in timers:
listening sockets run no timers.

show more ...

do not send FIN before end of data if options cause us not to send all data;
do not bother to send an empty segment if we cannot allocate an mbuf for data.

fix nasty bug that can cause memory corruption:
cannot use M_WAITOK in in_pcballoc calls for incoming TCP connection requests;
current kmem_malloc code can remap CMAP area, reentering clearseg or
phy

fix nasty bug that can cause memory corruption:
cannot use M_WAITOK in in_pcballoc calls for incoming TCP connection requests;
current kmem_malloc code can remap CMAP area, reentering clearseg or
physcopyseg with nasty effects. Could check SS_NOFDREF and use M_NOWAIT
in that case only, but that seems iffy.

show more ...

Always check the checksum on incoming packets that have them,
redefining udpcksum to control only generation of checksums

missing ifdef MROUTING (from hibler)

pass valid mbuf to in_cksum (from mccanne@horse.ee.lbl.gov)

back out revision 8.9 (re karels)

delete unused prototype

update for SETOPT/GETOPT (from karels)

set TF_ACKNOW rather than TF_DELACK when TH_PUSH bit is set (from mycroft)

arpwhohas prototyped twice

add a statistic that netBSD keeps

Fix two TCP bugs. The first is to recognize and
generate RST packets in response to packets that arrive on
a LISTEN socket that have the ACK bit set. The second fix
is to make sure that when we rec

Fix two TCP bugs. The first is to recognize and
generate RST packets in response to packets that arrive on
a LISTEN socket that have the ACK bit set. The second fix
is to make sure that when we receive a duplicate SYN/ACK
packet that we always generate an ACK. (The code that deals
with crossing SYNs broke this...)

show more ...

wrong sense of comparison (from karels)

make return types correct for tp functions

casts to quiet compiler

ensure that the option mbuf is freed (from mycroft)

ioctl cmd arg is u_long

off by one error prevents installing routes for class A interfaces
(fix provided by Karl Fox of Morningstar, reported via karels).